Rami Maalouf

November 18, 2024

Introduction (Executive Summary)

This report explores the creation and evaluation of a Generative AI (GenAI) prompt aimed at making the journaling process more intuitive and accessible. By leveraging the capabilities of large language models (LLMs), the prompt seeks to remove obstacles to reflection, uncover deeper insights from personal experiences, and seamlessly organize these reflections into a structured format. The approach focuses on transforming raw, unstructured thoughts into meaningful narratives, integrating user transcripts with a Markdown-based journaling system to bridge the gap between introspection and actionable outcomes.

Tools and Models

The prompt is intended for use with text-based or text-capable Large Language Models.
Testing was mainly conducted with agenta.ai cloud (cloud.agenta.ai) using the GPT-4 API

Background and Motivation

I am a true believer that within us we are able to come up with profound learnings that we can use to elevate our lives for the better. These learnings in my opinion are more valuable than any book or content on the internet because it comes from personal experiences so lessons get ingrained In us unlike just reading about other people’s experiences
- As Paulo Coelho said: “People never learn anything by being told, they have to find out for themselves.”¹
One of the best ways to extract knowledge And learn from your personal experiences is through journaling.
Unfortunately for journaling to work we are required to sit and deeply reflect. Whether it’s reflecting on our day, week, month, year, or even life.
- In this day and age, it’s becoming harder and harder to spend some time to sit and reflect.² So for that reason, for the past few months, I’ve been on a mission to find the best ways to leverage GenAI to reduce the barrier for me to not only enter a reflective “flow” but also capture that information in some way and store it in my knowledge base
I personally use obsidian to store my knowledge base and do my journaling in. For that reason, my ultimate goal with this assignment and prompt is to use it In my Obsidian plugin/extension that I will be building

Illustrative Use Cases or Examples

There are several possible use-cases for this prompt. On a high level, all my prompt is doing is:

taking in some messy input →
cleaning it up →
storing that information into the right place to increase the chances of future retrieval

Each step serves a purpose:
1 → 2:
- Whatever way you choose to express your thoughts (In my case through talking to myself). The prompt will then take that text that might not be the most accurate and clean it up (or even enhance it by generating some actionable takeaways based on the information I shared). It’s like taking a handful of raw ore and revealing the gold hidden within.
- studies have shown that these tools can assist students in brainstorming ideas, evaluating their learning strategies, and reflecting on their design processes, thereby deepening their self-awareness and understanding of their cognitive processes³
2 → 3:
- Now the question is: How can I leverage that information and use it in my own life. We need to store that information somewhere where it would be easy and appropriate for me to rediscover that information.

Examples

I will be using my Markdown file that I use for journaling for the 2 examples below.

tags:
- reviews/daily
  Created: 2024-11-04T15:16:36
Headings:
- "\[\[2024-11-04#Thoughts|💭\]\] \[\[2024-11-04#Improvements|💪\]\] \[\[2024-11-04#Obstacles|🚧\]\]"
- "\[\[2024-11-04#Accomplishments|✅\]\] \[\[2024-11-04#Gratitude|🙏\]\] \[\[2024-11-04#Content Log|📚\]\]"
  Parent: "\[\[My Calendar/My Weekly Notes/2024-W45|2024-W45\]\]"
journal: Day
journal-start-date: 2024-11-04
journal-end-date: 2024-11-04
journal-section: day
Dreams:
Summary:
Mindfulness:
Discipline:
Engagement:
Focus:
Courage:
Authenticity:
Purpose:
Energy:
Communication:
Uniqueness:
Rating:
---
 
## Cleaned Transcript
 
## Reflection
 
### Accomplishments
 
 
### Obstacles
 
 
### Thoughts
 
 
### Conversations

Example 1

This is a transcript from one of my journaling sessions that I had on November 4th. When I started speaking, I didn’t really have much to talk about, but as I continued to reflect on my day, I came up with more interesting realizations that I would have never came up with if I didn’t spend an extended amount of time reflecting on my day. It’s usually very hard for me and most people to dig deep into their thoughts but when all you have to do is speak to yourself (while doing something else like walking, chores, etc…)

Speaker 1 00:00 was not really a crazy day, so I’d say I would give it. Four in uniqueness. Uh, but nonetheless it was still a good day. Overall, I’d say it’s a six. Um, in terms of conversation, I was not. The best. I noticed. Um, Sometimes, I’m noticing recently that I’ve been doing. 00:31 I’ve been having. A little more small talk. Uh, and I would like to have Um, nothing wrong with it. I just think that I’d like to slowly. Escape that small talk. Especially if I’m having a Long conversation with someone. So, my focus for for tomorrow would be Uh, having more mindful and And slowly. 01:04 But surely. Move away from. Surface level conversation. 01:14 So after that, say I’ll give myself a Because I did manage to have a really good conversation. Uh, despite it being very short. Uh, with luck. 01:31 We talked about coaching. And, And our future plans and how we both aspire to. What we’re good at, and Teach people. Because at least for me, I Find a lot of satisfaction in that. And, Joy, and also, The people that I see who do that. I have so much respect for them. And for me, When it comes to, when I realized that, I have some respect for someone, It means that. I want to be. Like them in some way. So, I kind of Narrow down and focus and assess. What is That. Thing that I respect them for. I was generally able to focus. So I’d give myself a six today. But unfortunately, I couldn’t get Big progress done today. I was just stuck on one problem that Was that still haunting me? For the past two days. I spent. In total around. 17 hours. On this problem. Just this week, just the past three days. But I also spent a lot of time. Like, a few months ago when I was stuck on it. So, That made me kind of think about, The. What I want to do in my life. Kind of ignited. My. 03:29 Crisis of seeing. Reflecting on. If I really want to do this for the rest of my life, And this is where. Coaching came up. 03:47 And the rest is history. In terms of mindfulness, I’d say It was pretty good. Maybe a six as well. Um, 04:06 Regarding courage, I’d say. F7. Um, Because specifically, just for Act that I did. Which was? I saw this girl. That I thought was very cute. And she had a skirt. She was wearing a skirt. Which is not something. That is weren’t a lot these days. 04:43 Worn a lot. 04:48 So, I thought to myself, like, I, I had so much resistance to actually talking to her. Because as, you know, me, I Struggle to If I don’t have a reason, 05:08 To talk to someone. A valid reason. A logical one. 05:17 I face so much resistance. To talking to them because I do not want. Come off as creepy or trying to get something out of them. 05:31 So I basically dug Reasons to talk to her. 05:40 And that’s where I saw the skirt that she was wearing. I was like, okay, I guess I will compliment her. 05:52 And, Color. 05:58 Oh, the skirt. Which was that greenish? Bluish kind of color, but Speaker 1 06:07 Not turquoise. Speaker 1 06:18 But yeah. I ended up not actually. 06:25 Saying, anything else I just complimented her as I was passing by her, And then she left the dining hall. So, Yeah, I was just Is just a one. Or. 06:45 Quick interaction. But if I ever see her again, I guess. Now I 06:59 I’m no longer a complete stranger. 07:05 So, we’ll see where that goes. 07:10 And that kind of makes me think of the fact that Speaker 1 07:14 My mindset of. Speaker 1 07:16 Or my mindset on how I Talk to women has drastically changed. Just this past five days. 07:29 Um, Being more specific basically what changed? My mindset on. When to talk to them. And, My reasoning behind talking to them. Has changed. And as a result, I’m now more Open. Talking too much more. A lot more. 07:59 Than before. And, 08:07 It has honestly been 08:13 Very energizing. 08:20 Allow myself. 08:26 Talk to a lot of women. 08:37 As I mentioned before my caffeine, Social is social interactions. 08:46 It gave me so much energy when I socially interact with someone and talk to someone, especially if it’s like some sort of deep conversation. 08:58 But what I’ve noticed now is, Social interaction with a woman specifically. 09:09 Kind of, Amplifies the effects. 09:17 And, The more I talk to women. The more, I Have confidence in talking to more women. So, It’s a very 09:37 Good positive feedback loop. 09:44 It’s not only confidence in talking to women but also just confidence and everything. Like, after I talked to a woman and then I go Some some work or something. I go in there with a chip of my shoulder. 10:06 And I guess that’s how we operate. 10:15 Our hormones. 10:20 Do wild things. 10:27 Which, Leads me to. The importance. Of controlling these hormones and Not letting yourself. Do things that are not. Within your morals. 10:46 Or values. 10:51 Uh, so regarding today. My discipline score would be. 11:00 I was gonna give it a five but I’d say six actually. Just because I actually got myself. Uh, talk to that girl because I A small part of me would regret it if I did not compliment that girl. With a beautiful skirt. So I made myself do it. Just because I knew that I would draw credit if I did not. Regret it. And I didn’t really think of any reasons not to Approach. So I just did it. And regarding, My energy score for today. I would. It’s a little lower than yesterday because I did not talk to as much women. Uh, but it was still not bad. Let’s say maybe a six as well. 11:56 And then, Authenticity. 12:02 Maybe a five, because Mainly affected by my communication score. It’s because my communication score was low. Uh, authenticity is also low because of that. But other than that, It was not bad. See, authenticity is a five for today. And, 12:30 Regarding regarding engagement, which is basically. Focus. For everything other than work. I, Not sure how much I was engaged or not. I’d say just Then finally, the purpose. My purpose, I did not. Unfortunately, because I’m stuck on this stupid problem. That I’ve been facing for the past three days in the code. 13:05 It’s, it’s really mentally. Messing me up. And the result. Is. 13:19 Sense of purposeness. And what I’m doing. So, maybe 13:30 For purpose. 13:34 F3, a score of three. Speaker 1 13:39 So, yeah, for tomorrow I’d say My focus is having more meaningful conversations

Markdown output

tags:
- reviews/daily
  Created: 2024-11-04T15:16:36
Headings:
- "\[\[2024-11-04#Thoughts|💭\]\] \[\[2024-11-04#Improvements|💪\]\] \[\[2024-11-04#Obstacles|🚧\]\]"
- "\[\[2024-11-04#Accomplishments|✅\]\] \[\[2024-11-04#Gratitude|🙏\]\] \[\[2024-11-04#Content Log|📚\]\]"
  Parent: "\[\[My Calendar/My Weekly Notes/2024-W45|2024-W45\]\]"
journal: Day
journal-start-date: 2024-11-04
journal-end-date: 2024-11-04
journal-section: day
Dreams:
Summary:
Mindfulness: 6
Discipline: 6
Engagement:
Focus: 6
Courage: 7
Authenticity: 5
Purpose: 3
Energy: 6
Communication:
Uniqueness: 4
Rating: 6
 
---
 
## Cleaned Transcript
 
### General Overview
Yesterday was not really a crazy day, so I'd say I would give it a four in uniqueness. Nonetheless, it was still a good day overall; I'd say it's a six.
 
### Conversations
In terms of conversation, I was not the best. I've been noticing recently that I've been engaging in more small talk. Nothing is wrong with it, but I think I’d like to slowly escape that, especially if I'm having a long conversation with someone. My focus for tomorrow will be to have more mindful conversations and slowly, but surely, move away from surface-level conversation. Nonetheless, I did manage to have a really good conversation, despite it being very short, with Luck. We talked about coaching, our future plans, and how we both aspire to teach people what we're good at, as we find a lot of satisfaction and joy in that.
 
### Focus and Progress
I was generally able to focus, so I'd give myself a six today. Unfortunately, I couldn't make significant progress on a big problem that's been haunting me for the past two days. I spent around 17 hours on this issue just this past three days, and even more a few months ago. This situation has made me reflect on what I want to do in my life, and coaching came up as a part of that reflection.
 
### Mindfulness and Courage
In terms of mindfulness, I'd say it was pretty good, maybe a six as well. Regarding courage, I’d say seven. Specifically because I saw a girl who I thought was very cute, wearing a skirt, which isn't a common sight these days. I felt resistant to talk to her because I struggle to approach someone without a logical reason, fearing I might come off as creepy. I eventually found a reason — to compliment her skirt. It turned into a quick interaction, but if I ever see her again, I’m no longer a stranger. This changed my mindset significantly on how and why I talk to women. I've found it energizing and confidence-boosting to engage in such interactions.
 
### Engagement and Discipline
My discipline score would be a six. I approached the girl, knowing that I might regret it if I didn't. My energy was a little lower than yesterday because I didn't talk to as many women, but it was okay, around a six as well. Authenticity is maybe a five, mainly affected by my communication. Engagement, which is my focus on everything other than work, I’m not sure; I'd say just average. Finally, purpose is affected by my stuck problem in the code, so maybe a three.
 
### Improvements for Tomorrow
For tomorrow, I’d say my focus is having more meaningful conversations.
 
## Reflection
 
### Accomplishments
- Managed to have a short but meaningful conversation about coaching and future plans with Luck.
 
### Obstacles
- Struggled with a coding problem for 17 hours over the past three days without making significant progress. This has caused me to question my long-term career goals, introducing uncertainty about whether this is what I want to pursue for life.
 
### Thoughts
- Reflecting on moving away from small talk towards deeper, more meaningful conversations.
- Realizing a change in mindset regarding talking to women, finding it increasingly energizing and confidence-boosting.
 
### Conversations
#### Luck
- Discussed coaching, future aspirations, and mutual satisfaction found in teaching others.
 
### Improvements
- Focus on having more mindful and meaningful conversations rather than surface-level interactions.
- Implement a strategy to tackle persistent problems by taking breaks or seeking input from peers to avoid burnout

Example 2

This is a transcript from one of my journaling sessions that I had on November 5th as I was preparing myself to sleep after an eventful day that I want to write about so that I can look back at it one day in the far future.

Crazy day. So Finally. Got myself. To get out of my desk. And go out and talk to people and record myself. While doing that. Life. And I finally basically in other words,00:25Uh, switched to focusing on. Content creation. And honesty. I did not see that aside of University. During the night evening time. I didn’t know, it gets crowded like that.00:47So, I’ll definitely be. Going there more often.00:55I literally just went out of my house just because the weather is nice. And my only plan was To do something. Something, social1Speaker 101:08I1Speaker 101:09had a poster saying, free hugs. Because I couldn’t find a better thing to do. Um, So yeah, I just walked around. The plan was that just walk around and get free. Hugs. And through that add also document. Thought process.01:34It was weird. Because I don’t have I didn’t have a structure. I don’t have a plan. I was just dragged to whatever happens. So what happened is Smith? The guy who, Somehow broke my tripod. Came back from Best Buy. To get me one. Or he just got me one. So, I,02:08I had a bigger reason that I was getting. Things are starting. We’re starting to like. Work for me. So, And then also, he mentioned that Uh, he had a Vr headset, which I Basically spent the majority of my time.02:34So, What happened is? I started by wearing this. This VR headset and just walked around. Uh, asking people for hugs. I somehow. Find this group of Japanese exchange students. Who are just dying from laughter, especially one of them. That1Speaker 102:59that1Speaker 102:59kind of made my day. The fact that I made that little girl, Not even little, she’s just short.1Speaker 103:08Um,1Speaker 103:09Laugh. Hysterically. Lovely to see.03:20What else? Yeah, I just continued interviewing strangers. I did that. I talked to some Arabs. But yeah, regarding conversations though. It wasn’t the best I’d say, Those not bad. Let’s say F. A mid one. Nothing crazy. Well, the thing that I create that was crazy was My confidence. I think. I felt so confident. Actually, maybe not too much. I’d say an For today. It’s good. And then,04:11So yeah, for confidence I mean, for socialization. Um, it was a But,04:26The number five.04:30But then,04:34Uh, regarding my04:38Confidence. Of course I was at my Peak. I, Very stupid and random things. That I wouldn’t have done.04:54It was if I was thinking straight. For example, punching the air Extended amount of time.05:06Uh, And some random people. Put put it on their stories. And,1Speaker 105:33Yeah, like1Speaker 105:34this kind of shows. To me that. I have very little to no ego. I cannot. Or I don’t really care about how I see myself.05:53Like, I I looking from the outside after like, watching the videos. I look so stupid. But I didn’t think of any of that when I was doing it. I was just enjoying it. Giving people laughter. Whether they’re making fun of me or not, I don’t care.06:18But in general, it was It was just. Stupid. Like what I was doing is just stupid. But, I think it really helps just being comfortable with doing things like that are out of the social Norm, like punching the air while playing well well wearing a VR headset.06:45Here’s the thing, like that kind of really aligns. With. What I want to do, or what I strive to show people. I want to show people, I strive to show people.1Speaker 107:01That1Speaker 107:01the things they can do.07:06The things that they stop themselves to do. The things that they truly want to do.07:14People stop themselves from1Speaker 107:17Doing1Speaker 107:17things they really want to do. Because of, Social norms societal expectations, and all that stuff.07:29I guess. What my video could show.07:36To overcome that.1Speaker 107:43And just That’s normal.1Speaker 107:49Maybe that’s the point of the video that I would be making. Just showing people. Myself doing weird things. And then justifying how doing this is not weird. And how you could do it too.1Speaker 108:29So, regarding1Speaker 108:30purpose, I’d say Either a score. Of. Seven. It’s like I’d give it a seven. Regarding confidence. I’d say.So yeah, I don’t think I talked about my engagement level yesterday.00:08I think. I was quite engaged. I give myself a seven. For engagement. I was. At least trying to be very engaged since. Yeah, I have to because I’m doing a YouTube video.00:29Yeah, I think, as I mentioned previously, My conversations were not the best because I was not very Mindful, you can say.00:41Uh, but That was while recording the video. But as for the video, when I went to the philosophy Club event, I have that I went to after my video recordings. I had a had good conversations there and I was listening. Because they had really cool conversation topics. So I really enjoyed that01:10And then, So, yeah, uh, in terms of Engagement, I’ll give myself a seven just because of My ability to be engaged during that philosophy Club event. Which was really fun. Uh, what else?01:33Focus that I talk about Focus before. If I didn’t, I don’t know.01:43I just struggled to focus today to be honest. Which kind of helped with making me actually get out and do shit. Uh, but Yeah, I just found myself. Stuck with this assignment that I’m still working on. I don’t think so much resistance to actually doing it.02:11Yeah, I’m getting more motivation today, to do it. So, we’ll see how that goes today.02:25What else?02:31Our courage. I think. Yeah, I mentioned it was an For Carriage. Because, I did crazy stuff. I was punching the wall, the air. Acting like I’m Superman.1Speaker 102:51And1Speaker 102:51at some point, I even got punched myself. In the air. And I also beat my arch nemesis. In the So, It’s basically it was basically improv. So I it was interesting.03:14And, Yeah, I hope I Inspired someone in some way. But yeah, one one key. Factor one, key area. That I would like to focus on. Is that person telling me? Is this a club?03:40And that made me remember. What my friend told me about how I should really strive to. Build the club out of this and make That.03:55Aims. Help people. Overcome their shyness.04:06By doing things uncomfortably. Doing uncomfortable things. Facing their fears.04:15Maybe specifically, for my case, for my club, it would be facing people’s social anxieties. Honestly, that shit would mean so much to me that club. I would love to Lead, that kind of Club.04:36So yeah. I just really don’t know how to execute it. I really have to. Talk to people and see what they really want and see what I really want from this. And how I can provide the most value. To people struggling with what I used to struggle with.05:07It’s truly something that I really want to do. Uh it’s just uh the hardest part is getting to understand what it what is really needed? From the people around me from the world from University from the University. What does the University need from?05:40I guess for starters, I was thinking, like, Since I do not want to make it about the money for now. But, So yeah, I’ll make it free but donations are always welcome. But, I would essentially.06:00Coach people are there. Or basically form a group. Where I go with them, go out with them and We help each other. Make. Crazy challenges. Deuce crazy shit. Just to get outside of our heads. And, Live in a moment. And overcome. Their fear of social anxiety.06:39Um, you can see, I’m kind of trying to be Trying to be an exposure therapist. Which I’m not certified to be.

Markdown output

tags:
- reviews/daily
  Created: 2024-11-05T11:54:44
Headings:
- "\[\[2024-11-05#Thoughts|💭\]\] \[\[2024-11-05#Improvements|💪\]\] \[\[2024-11-05#Obstacles|🚧\]\]"
- "\[\[2024-11-05#Accomplishments|✅\]\] \[\[2024-11-05#Gratitude|🙏\]\] \[\[2024-11-05#Content Log|📚\]\]"
  Parent: "\[\[My Calendar/My Weekly Notes/2024-W45|2024-W45\]\]"
journal: Day
journal-start-date: 2024-11-05
journal-end-date: 2024-11-05
journal-section: day
Dreams:
Summary: Today, I focused on content creation through social interaction experiments. I realized the need to challenge social norms and perhaps form a club to help others overcome social anxiety.
Mindfulness: 5
Discipline:
Engagement: 7
Focus: 4
Courage: 8
Authenticity:
Purpose: 7
Energy:
Communication:
Uniqueness:
Rating: 6
 
---
 
## Cleaned Transcript
 
### Content Creation and Social Experiment
 
Today, I finally got myself to leave my desk and interact with people, recording myself while doing so. I shifted my focus to content creation. To my surprise, the university was crowded in the evening, something I hadn’t realized before. I plan to visit more often.
 
I stepped out of the house because the weather was nice, with no solid plan apart from doing something social. I made a poster that said “free hugs” because I couldn’t think of anything better to do. I walked around, collecting free hugs, documenting my thought process, and interacting with strangers.
 
### Memorable Interactions
 
While my day lacked structure, meeting a guy named Smith was a highlight. He managed to break my tripod but later returned with a new one from Best Buy. This gesture made me feel that things were beginning to align for me.
 
I spent most of my time with Smith using a VR headset, where I also approached people for hugs. I encountered a group of Japanese exchange students who were amused by it, especially one girl who laughed hysterically. Making her laugh was a highlight of my day.
 
### Reflections on Confidence and Social Norms
 
Interacting with others boosted my confidence. I felt incredibly confident, perhaps at a peak, doing random and unconventional things, l like pretending to punch the air while wearing a VR headset. Videos of me doing this circulated on social media, showing that I lack ego regarding people's perceptions of me. Watching these videos later made me realize how odd I looked, but at the moment, I didn’t care, focusing instead on spreading joy.
 
This experience aligns with my intention to show people they can do things outside societal norms and expectations. My actions and videos aim to encourage others to overcome such limitations.
 
### Personal Goals and Club Ideas
 
Regarding my sense of purpose, I rate it a seven. My engagement was high, particularly during a Philosophy Club event, so I’ll give myself a seven in that area as well. My focus wavered, but it motivated me to step out. Confidence was high, manifesting as courage to act unconventionally. I punched the air and enacted scenarios of beating an arch-nemesis. Courage deserves a high rating for today's antics.
 
I had a meaningful conversation where someone suggested turning these social interactions into a club. This reminded me of a friend's advice to create a platform helping people overcome shyness and social anxiety by confronting uncomfortable situations. This concept excites me, though I need to understand what people need and how I can deliver the most value. The club would help people deal with social fears, something meaningful to me as I have overcome such struggles.
 
## Reflection
 
### Accomplishments
- Successfully shifted focus to content creation and engaged with people
- Managed to organize a social experiment with a "free hugs" poster
- Interacted with a group of Japanese exchange students, bringing joy
- Attended a Philosophy Club event and participated in meaningful conversations
 
### Obstacles
- Felt unstructured throughout the activities
- Struggled with focus and had resistance to complete assignments
 
### Thoughts
- Consider the idea of forming a club to help others confront their social anxieties
- Reflect on how unconventional actions can inspire confidence and freedom
 
### Improvements
- Develop a clearer structure for future social experiments to enhance effectiveness
- Work on completing assignments to avoid procrastination
- Explore practical steps to create a supportive club environment
 
### Conversations
- Discussion with Smith led to positive outcomes, showing potential value in collaborative efforts
- Reflections on interactions at the Philosophy Club provided insights into potential club setup

Research Literature Informing the Prompt

The writing style of a wise advisor, mentor, or life coach is characterized by several key elements:

Clarity and Simplicity: They use straightforward language to ensure their messages are easily understood.
Empathy and Understanding: Their writing reflects a deep comprehension of the reader’s experiences and emotions.
Encouragement and Positivity: They focus on uplifting and motivating the reader, highlighting strengths and potential.
Guidance and Direction: They provide actionable advice and clear steps to help the reader achieve their goals.
Reflective and Thought-Provoking: They pose questions or insights that encourage self-reflection and personal growth.
Authenticity and Sincerity: Their writing conveys genuine care and a desire to support the reader’s journey.⁴⁵⁶⁷⁸

These elements combine to create a supportive and effective communication style that fosters development and empowerment. Making the model use this writing style will prime it to respond more effectively for the purpose of providing the best advice given any situation

Key Objectives

Automate Reflection: Convert raw transcripts into cleaned, structured content.
Generate Insights: Extract and enhance key learnings and takeaways.
Streamline Knowledge Storage: Organize insights into an easily retrievable Markdown format for future reference.

In the following sections, we use the above list of characteristics as the backbone of our prompt engineering approach. The above list was intentionally made in a concise manner to summarize a large amount of information, so that the prompt could encapsulate a great amount of instruction without using too much of the context window.

Outline of Prompt or Prompting Technique

Starter prompt:

Given this transcript: {transcript} Please analyze it and improve on it and write down some actionable takeaways based on the transcript. I then want to place that information into my Obsidian (markdown) file which looks like this: {md-file}

Prompt Engineering Tests and Evaluation

Testing was mainly conducted with agenta.ai cloud (cloud.agenta.ai) using the GPT-4 API. The angeta.ai platforms allows for definition of variables in the prompt in order to test variations of the prompt or to apply it to varying input texts. In the following prompts, items that are treated as variables by angeta.ai are displayed in brackets {} similar to the agenta.ai syntax.

I tested three variations of the prompt (vanilla, detailed description, and detailed description plus multi-shot). The detailed description in the second prompt variant was prepared based on the research outlines in section 5. The five examples used for the multi-shot prompt were generated with the o1-preview model from OpenAI after being given the detailed description in the second prompt variant.

Evaluations of these prompts were conducted on the three different source texts mentioned above (the quotes from Malthus, Mill, and Davis) each from scholarly literature but containing important ideas in the history of thought. I also tested a fourth text which was the abstract of an article I have published myself, because of my own deep understanding of the source text which would give me a deeper appreciation of the quality of the style transfer done by the LLM. This fourth source text was as follows:

“By comparing the ideal-type theory of strategy without entrepreneurship with the ideal-type theory of entrepreneurship without strategy, I shed light on what it means to combine the logics of entrepreneurship and strategy and illustrate the value of their complementarity. The key insight is that entrepreneurship is blind without strategy and strategy is paralyzed without entrepreneurship. I further argue that many proponents of strategic entrepreneurship and action-based theories of strategy have prematurely given up on the idea of sustained competitive advantage. Disequilibrium and uncertainty do not automatically mean that no predictable and stable patterns can be relied upon for sustained competitive advantage.”[26]

Each of the above four original texts were repeated three times in the test set for agenta.ai in order to produce more data and reduce the likelihood of chance outcomes influencing the overall averages. So my test set had a total of 12 rows and 2 columns (see screenshot below).

The method of evaluation was to use the LLM-as-a-judge custom evaluator feature in agenta.ai with three different evaluation prompts (simple 0-10 scale, 0-10 scale with detailed description and instructed to be a tough grader, and lastly a 0-10 scale extremely tough grader with no detailed description).

My initial results showed that getting the evaluators to be tough graders was hard and they were inclined to be very positive in their assessments. I had to revise the wording to get the evaluators to be tougher on grading.

Each of the prompt variants tested and the evaluation prompts used to test them are provided below.

Agenta.ai settings used for tests

Prompt system: I left this field empty

Prompt user: One of the prompt variants listed below

Model parameters: I mostly kept the defaults but changed the model to the most powerful one available (gpt-4). Temperature: 1, Max tokens: -1, Top p: 1, Frequence penalty: 0, Presence penalty: 0, Force json: no.

When creating a new evaluation to test all three prompt variants with our LLM-as-a-judge evaluator, I set the “Batch Size” parameter to 1 in the advanced configuration settings.

Note that in addition to the “md-file” and “transcript” column in my Test Set, I had to add a “correct_answer” column too or else the evaluator would return an error. I put the text “There is no right or wrong answer as this is a subjective task” in the “correct_answer” column for all test cases.

Prompt variant 1 (vanilla):

Prompt variant 2 (detailed):

You are a reflection expert and my personal wise advisor who has extensive knowledge in reading my thoughts and reflections. Your one and only goal is to go through 2 consecutive steps/pipelines that helps you fulfill your role as a reflection expert and my personal wise advisor.

Step 1: Every day, I will be giving you a transcript of my daily reflection where I go through how my day went and share any key memorable moments, what I learned, and improvements for tomorrow. Your goal for this step is to take transcript, which might contain transcription inaccuracies, and correct these transcription-induced errors to the best of your abilities while following these guidelines:

Fix common typographical errors, including but not limited to spelling mistakes, misuse of punctuation, incomplete sentences, and improper capitalization.
Use context and common sense to correct errors
Only fix clear errors, don’t alter the content unnecessarily
Maintain a similar writing style as the way I speak, which is first person, and
Group my transcript into different headings, each one tackling a different topic

Step 2: After cleaning and polishing the transcript, please take some time to deeply reflect and analyze that transcript. You goal for step 2 is to take this cleaned transcript and populate my daily journal markdown file that you can take a look at here

+++ START OF DAILY JOURNAL MARKDOWN FILE

{md-file}

+++ END OF DAILY JOURNAL MARKDOWN FILE

As you can see, the markdown file contains several inputs in the data and several questions within the file content. To help achieve the ultimate goal for step 2 fully and correctly. We will split step 2 into 2 tasks:

Task 1:
- One of the headers that you will find is called “Cleaned Transcript”. In there, please place the cleaned transcript that you wrote during step one.
Task 2:
- After you ensure you are fully done with task 1, please leverage that cleaned transcript to populate the rest of the file.
- Start with populating the metadata keys like “Summary” and all keys that lie under that. For these keys under the “Summary”, please fill them with a number between 0 and 10, based on the information that I shared in the transcript where I mentioned a certain number for each key in the metadata. So for example, I might say something like “thanks to my good sleep I felt so energetic so I would give myself an eight on energy”. This means that for the key called “Energy”, you would fill it with an “8”. Do the same thing for all the metadata fields that lie under “Summary”. I will most likely share a number for each of the rating metadata but if none was mentioned in the transcript, please leave that metadata key empty.
- After filling out all the metadata, the next step would be filling out the headers and questions that are in my markdown file template
  - for example, one of the headers in the template is “Improvements”. What you need to do for this is deeply reflect on the cleaned transcript and look for things that I mentioned throughout the transcript where I signify that I would like to get better at something. You can also place a few of your own that you would recommend based on your analysis of my transcript as a personal wise advisor Alright let’s begin!

+++ START OF TRANSCRIPT TO USE FOR STEP 1

{transcript}

+++ END OF TRANSCRIPT TO USE FOR STEP 1

+++ START OF UPDATED MARKDOWN FILE

Prompt variant 3 (detailed and follows prompt crafting framework):

You are a reflection expert and my personal wise advisor who has extensive knowledge in reading my thoughts and reflections. Your one and only goal is to go through 3 consecutive steps/pipelines that helps you fulfill your role as a reflection expert and my personal wise advisor.

Fix common typographical errors, including but not limited to spelling mistakes, misuse of punctuation, incomplete sentences, and improper capitalization.
Use context and common sense to correct errors
Only fix clear errors, don’t alter the content unnecessarily
Maintain a similar writing style as the way I speak, which is first person, and
Group my transcript into different headings, each one tackling a different topic

+++ START OF DAILY JOURNAL MARKDOWN FILE

{md-file}

+++ END OF DAILY JOURNAL MARKDOWN FILE

Task 1:
- One of the headers that you will find is one called “Cleaned Transcript”. In there, please place the cleaned transcript that you wrote during step one.
Task 2:
- After you ensure you are fully done with task 1, please leverage that cleaned transcript to populate the rest of the file.
- Start with populating the metadata keys like “Summary” with a summarized version of the cleaned transcript. Then populate all keys that lie under “Summary” with a number between 0 and 10, based on the information that I shared in the raw transcript where I mentioned a certain number for each key in the metadata. So for example, I might say something like “thanks to my good sleep I felt so energetic so I would give myself an eight on energy”. This means that for the key called “Energy”, you would fill it with an “8”. Do the same thing for all the metadata fields that lie under “Summary”. For “Rating”, which is just my overall score for the day, I might say something like “overall, I would say today was a five”, so that means you will need to populate the “Rating” with “5”. I will most likely share a number for each of the rating metadata but if none was mentioned in the transcript, please leave that metadata key empty. All these metadatas under “Summary” should be filled with a number, anything else would break the markdown file.
- After filling out all the metadata, the next step would be filling out the headers and questions that are in my markdown file template
  - for example, one of the headers in the template is “Improvements”. What you need to do for this is deeply reflect on the cleaned transcript and look for things that I mentioned throughout the transcript where I signify that I would like to get better at something. To also fulfill you’re role as my wise advisor, please place a few of your own advice for improvements that you would recommend based on your analysis of my transcript. Ensure the advice and suggestions for improvements are practical, personalized, and connected to a deeper reason/motivation. Alright let’s begin!

+++ START OF TRANSCRIPT TO USE FOR STEP 1

{transcript}

+++ END OF TRANSCRIPT TO USE FOR STEP 1

Please feel free to write your thoughts out as you go through the different steps, but when it’s time for you to rewrite the markdown file, I need you to give me the ABSOLUTELY COMPLETE revised markdown without ANY new headers and without adding or removing any metadata

Evaluation Prompt (0-10 scale):

System prompt

You are an evaluator grading an LLM App.

You will be given a RAW TRANSCRIPT, DAILY JOURNAL MARKDOWN FILE, and WISE ADVISOR’S THOUGHTS AND UPDATED MD FILE.

Here is the grade criteria to follow:

Ensure that the LLM APP OUTPUT has the same meaning as the CORRECT ANSWER

SCORE:

The score should be between 0 and 10
A score of 10 means that the answer is perfect. This is the highest (best) score.

A score of 0 means that the answer does not any of of the criteria. This is the lowest possible score you can give.

ANSWER ONLY THE SCORE. DO NOT USE MARKDOWN. DO NOT PROVIDE ANYTHING OTHER THAN THE NUMBER

User prompt

My wise advisor was given the task to take a transcript of my daily reflection where I go through how my day went and share any key memorable moments, what I learned, and improvements for tomorrow. The wise advisor is supposed to take that transcript and a template Markdown file that my wise advisor is supposed to fill it up. Your task is to evaluate and rate how well the wise advisor did.

+++ RAW TRANSCRIPT

{transcript}

+++ DAILY JOURNAL MARKDOWN FILE

{md-file}

+++ WISE ADVISOR’S THOUGHTS AND UPDATED MD FILE

{prediction}

+++ EVALUATION CRITERIA

My wise advisor is supposed to take a transcript, which might contain transcription inaccuracies, and correct these transcription-induced errors to the best of their abilities. 4 points of the final grade is from how well they were able to ‘clean’ the transcript Another 4 is based on how well they were able to extract key information from the transcript and place that information into the right places in the markdown file. So for example, if in the transcript I mention “overall, I would say today was a five”, so that means you will need to populate the “Rating” with “5”. The final 2 is based on how well they were able to update the provided markdown template while still maintaining the structure (Headers and metadata)

+++ EVALUATION INSTRUCTIONS

Evaluation strategy: 0 to 10 where 0 is very bad and 10 is very good.

The most important thing about your grading is that you have to be extremely tough. Mostly try to give low grades unless the expert REALLY deserved it.

Answer ONLY with one of the given grading or evaluation options. Just a single number. Do not give any additional information or context. Just give me a single number between 0 to 10. Make sure your response contains nothing but this number.

Results

The test results for each prompt variant was downloadable as a csv file from agenta.ai. The total cost of running the prompt engineering experiments across all three variants was $1.5818 (s u m o f t h e " T o t a lC os t " co l u mn) . T hi s i s l ik e l y in U S d o ll a rs . A m ore a cc u r a t ecos tw o u l d b e a ro u n d$ 3 but I deleted the failed evaluations and outdated evaluation prompts.

I used ChatGPT’s code interpreter to analyze this data. I started the analysis with the following prompt in ChatGPT with the model GPT-4o and code interpreter enabled using the “CSV Data Analyzer” custom GPT:

There are 3 different variants that I’m trying to compare: - “app.default v2” - “app.better v4” - “app.best v2” They are each a prompt engineering experiment on three variations of a prompt that aims to assess the performance of each variant to determine the relative performance of the three variants of the prompt and if there is any statistically significant difference between them. Let’s start by understanding the file structures, grouping the variants, and then selecting the best methodology to visualize the data, and test for statistical significance, and visualize the test results in beautiful data visualizations. I am not satisfied to know that just one of the prompt variants performed better. I want to know which one performed best, which one worst, and which one was in the middle. I also want each prompt variant tested against each other variant in a 1vs1 test on all three evaluation scales. I want the results of all these tests, along with their statistical significance at the 0.1 level (90%), 0.05 level (95%) displayed nicely in an aggregate table. After that, I would like the best visualization that captures these differences in your opinion.

After this I had to do multiple follow ups to get ChatGPT to provide the results in the format that I wanted.

Key Findings:

Variant 2 had the highest mean of 7.46. Followed by variant 3 with 7.38, then 6.63 for the base variant
It took an average of 18 seconds for GPT-4o to send the full response, variant 2 took 6 more seconds (24 s), and variant 3 another 6 (30 s).
The average cost was 0.028 for base variant, 0.050 for variant 2, 0.054 for variant 3 (almost double the cost of the base variant)

Were the results between the 3 different variants statistically significant?

Base variant vs. variant 2: Statistically significant difference at both 90% and 95% levels, with app.better v4 performing better. The p-value was 0.0037
Base variant vs. variant 3: Statistically significant at both 90% and 95% levels, with app.best v2 outperforming app.default v2. The t-value was 0.013
Variant 2 vs. variant 3: No statistically significant difference since the p-value was 0.70

Performance comparison chart


Prompt 1	Prompt 2	T-statistic	P-Value	Significant at 0.1	Significant at 0.05
base variant	variant 2	-3.32	0.0037	True	True
base variant	variant 3	-2.71	0.013	True	True
variant 2	variant 3	0.38	0.70	False	False

I validated an triangulated the data analysis of ChatGPT with code interpreter through a number of methods:

I reviewed the code and data as I myself have some expertise in statistical data analysis and programming.
I tried the same analysis again in a different instance of chat with ChatGPT (using both the base ChatGPT and the custom GPT “CSV Data Analyzer”)

Variations or Follow-ups

This prompt is only a step in the multi-step process of automating and reducing the barrier to reflecting and journaling.
- one way to expand on it would be splitting the process into several steps with several agents to improve the quality of the response
another idea could be: instead of sending over a transcript, we could use the realtime speech to speech model so the model can ask clarifying questions based on what needs to be filled in the file. To make the whole process interactive and personalized.
- this process can be applied to any other thing that is hard to get yourself to write down. Like an assignment

Usage considerations

This prompt was engineered by someone who is not an expert in:
- Journaling
- Reflection
- Psychology
- Mental health
An expert psychologist and prompt engineer would undoubtedly do a better job in the kind of analysis and prompt design attempted here, and would have a much better understanding of the existing research relevant to the prompt. As a non-expert, I may have made mistakes in how I read or interpret existing research or how I give weight to the research that I read in a domain outside of my expertise.

Ethical concerns

As much as I would love for GenAI to completely understand me and my thoughts, our thoughts are way too complex and sometimes can definitely be misinterpreted by GenAI, which as a result, might provide advice that is not relevant to us. The danger might come in when you trust the model too much that you start taking all it’s advice seriously, or worse, replacing your therapist with an LLM when you need professional help tailored to your unique emotions and circumstances. While GenAI can offer guidance, suggest coping strategies, or provide a sounding board for your thoughts, it lacks the nuanced understanding and ethical accountability of a trained therapist.⁹
- Relying solely on an AI model could lead to overlooking critical issues or misinterpreting subtle emotional cues that require human expertise. It’s important to use GenAI as a supplementary tool rather than a replacement for human connection and professional care, especially when dealing with mental health challenges.¹⁰
Sharing personal information to LLMs might bring up several privacy concerns if you are using a model from the cloud like from OpenAI’s ChatGPT.¹¹

Discussion and Conclusion

Agenta.ai only allowed me to use GPT-4 for my tests and for my LLM-as-a-judge results. It is possible that with a different model like GPT-4o, o1, Llama, Claude or Gemini, I may have arrived at different results
I should have used some more research-backed prompts. It might have improved the quality of response or provided better and more research-backed advice.
A few of the requirements for the prompt couldn’t really be research-backed. For example, the prompt for the markdown file editing, I couldn’t find any research for it so I just followed the “Prompt Crafting Framework” and a lot of trial and error to finally get it right,

‌

The Change Companies. (2019, December 2). How Interactive Journals Promote Behavior Change. Changecompanies.net; The Change Companies. https://blog.changecompanies.net/interactive-journals-behavior-change ↩
Cai, I. (2016, May 9). The Power of Journaling in Learning, Mindset and Behavioural Change. Iriscai.com. https://iriscai.com/blog/2016/5/9/the-power-of-journaling-in-learning-mindset-and-behavioral-change-part-i ↩
Nguyen, H., & Nguyen, A. (2024). Reflective Practices and Self-Regulated Learning in Designing with Generative Artificial Intelligence: An Ordered Network Analysis. Journal of Science Education and Technology. https://doi.org/10.1007/s10956-024-10175-z ↩
Nieuwerburgh, C. van, & Oades, L. (2017). Editorial. Coaching an International Journal of Theory Research and Practice, 10(2), 99–101. https://doi.org/10.1080/17521882.2017.1355828 ↩
Templeton, N. R., Jeong, S., & Pugliese, E. (2021). Mentoring for continuous improvement in teaching and learning. Mentoring & Tutoring Partnership in Learning, 29(1), 1–5. https://doi.org/10.1080/13611267.2021.1902709 ↩
Models of Continuing Professional Development: a framework for analysis. (2024). Journal of In-Service Education. https://doi.org/10.1080//13674580500200277 ↩
Educative mentoring in sport coaching: a reciprocal learning process. (2022). Cambridge Journal of Education. https://doi.org/10.1080//0305764X.2021.1990860 ↩
Understanding mentoring. (2024). The Vocational Aspect of Education. https://doi.org/10.1080//10408347308003881 ↩
Higgins, O., Short, B. L., Chalup, S. K., & Wilson, R. L. (2023). Artificial intelligence (AI) and machine learning (ML) based decision support systems in mental health: An integrative review. International Journal of Mental Health Nursing, 32(4), 966–978. https://doi.org/10.1111/inm.13114 ↩
Balcombe, L. (2023). AI Chatbots in Digital Mental Health. Informatics, 10(4), 82–82. https://doi.org/10.3390/informatics10040082 ↩
Huang, J., Shao, H., & Chang, K. C.-C. (2022). Are Large Pre-Trained Language Models Leaking Your Personal Information? ArXiv.org. https://arxiv.org/abs/2205.12628 ↩

Rami's Universe

Explorer

Breaking Barriers to Reflection - A GenAI-Driven Approach to Journaling

Introduction (Executive Summary)

Tools and Models

Background and Motivation

Illustrative Use Cases or Examples

Examples

Example 1

Markdown output

Example 2

Markdown output

Research Literature Informing the Prompt

Key Objectives

Outline of Prompt or Prompting Technique

Starter prompt:

Prompt Engineering Tests and Evaluation

Agenta.ai settings used for tests

Prompt variant 1 (vanilla):

Prompt variant 2 (detailed):

Prompt variant 3 (detailed and follows prompt crafting framework):

Evaluation Prompt (0-10 scale):

System prompt

User prompt

Results

Key Findings:

Were the results between the 3 different variants statistically significant?

Performance comparison chart

Variations or Follow-ups

Usage considerations

Ethical concerns

Discussion and Conclusion

Graph View

Table of Contents

Rami's Universe

Explorer

Breaking Barriers to Reflection - A GenAI-Driven Approach to Journaling

Introduction (Executive Summary)

Tools and Models

Background and Motivation

Illustrative Use Cases or Examples

Examples

Example 1

Markdown output

Example 2

Markdown output

Research Literature Informing the Prompt

Key Objectives

Outline of Prompt or Prompting Technique

Starter prompt:

Prompt Engineering Tests and Evaluation

Agenta.ai settings used for tests

Prompt variant 1 (vanilla):

Prompt variant 2 (detailed):

Prompt variant 3 (detailed and follows prompt crafting framework):

Evaluation Prompt (0-10 scale):

System prompt

User prompt

Results

Key Findings:

Were the results between the 3 different variants statistically significant?

Performance comparison chart

Variations or Follow-ups

Usage considerations

Ethical concerns

Discussion and Conclusion

Footnotes

Graph View

Table of Contents