Week 1 Recap: Ideas, Boilerplate, Monitoring & (not) Shipping
Here's what I accomplished in my first week of my 3-month build & learn cycle
I've been toying with LLMs since end of 2022, trying to create an AI Assistant for my company, Canopy. The tech wasn’t quite there until GPT-4 came along. Last week, after launching our AI Assistant in alpha, I decided to leave my job to focus on developing my own ideas. I’ll be documenting my journey here as I go, and this is the recap of my first week.
Why now?
With the last couple of years' progress, so many of the traditional ML projects have been (or could be) replaced by language models, making things easier compared to just three years ago. Things like unstructured data, classification, sentiment analysis, NLP, speech, and much more are now all manageable by one person or a tiny team.
I think there are four ways to benefit from what is happening:
Big Tech working on foundational infrastructure or large problems.
Moonshot projects challenging existing players like Perplexity vs Google.
Vertical applications (or what people are now calling "Applied AI").
Indie hackers focused on things that don't look venture-scale at first glance.
For now, for these next three months, I'm personally interested in item 4. Maybe later I'll try to join a team working on the other ones.
Setting a 3-month deadline
Having a short deadline helps keep me motivated and accountable. It pushes me to focus on practical, achievable goals.
Since I work in product and not ML or research, I'm interested in end-use application. There's a ton of effort involved in making LLMs (unstable, unreliable & expensive tech) work smoothly in a finished product.
For instance, if I were only studying, factors like accuracy, cost, or user experience would be glossed over. But if I need to ship and charge for something, that’s the opposite.
Three months, with my skill set, I think is a doable but still challenging timeline to ship at least two products (my stretch goal is six).
Now, to the first-week recap
I had 5 goals for this past week:
Ship my personal site, start a Substack, clean up social media.
Get a boilerplate codebase for my projects (sign-up, subscription, etc).
Organize and research my ideas: pick 3 to 6.
Canopy: monitor the Alpha launch of our AI Assistant.
Bonus: Ship a WIP project that I had already.
✅ Goal 1: Personal site, Substack & social media
Personal site: Over the weekend, I started working on a personal site that I’d need to send to some people. I also used it to play with Next, Vite, Astro, Tailwind, and Shadcn. All things not part of Canopy stack.
I went with Next + TailwindUI. Site is live here: https://daniellopes.dev
For my products will be using just Tailwind and my usual Rails.
Substack: Working with the garage door open is something I always wanted to do for Canopy, but I left that for my co-founder. So, last week I started this, and I posted the first post: Using LLMs to parse 7 years of stand-up updates (which was an exercise in trying different summarization approaches as well).
Writing a ton is also part of one of my ideas that I might work on next month.
Social media: To stay more informed about things happening in San Francisco (meetups, hackathons, etc). So spent a few hours making list of people I want to follow more closely. Social media isn't all just noise; it can be valuable.
🟨 Goal 2: Organize all my ideas
I made a lot of progress on this goal, but not enough to confidently have three solid ones. I'll keep working on this next week. Here's the gist:
I keep a list of ideas, logging anything that comes to mind (features or products). I always have anywhere around 100 to 200 items.
This week, primary goal was to organize my list, considering these 4 criteria:
Is it a newly viable thing? And is it flooded with people doing good work already? (Areas like Knowledge Base for Companies or Customer Support are already saturated.)
Do I want this for myself? Building something for others demands a lot of research, which isn't feasible within a three-month timeline.
Does it have commercial potential?
If no commercial potential, the aim should be for learning or personal use; don't waste time on unviable or tarpit ideas. Do just enough to meet the goal and move on. At the end of the day, I want people to use the things I build.
If there's commercial potential, do I have a unique perspective on solving the challenges? Then do the usual analysis—pricing, customer base needed for a MRR, competitors, etc.
Does it have the potential to find a single feature that could be a lot better than existing methods? And can it be developed as an add-on for existing players or does it need to be a replacement? (replacements are less interesting with this timeline).
What’s the marketing effort? Can I market it alone? If it's B2B, can I identify a sustainable group of people to target? If it's B2C, can I come up with marketing strategies that don't rely on ads or require excessive time?
Many ideas I had on the list from last year had no credible competitors six months ago, but now have promising ones. So, I ditched those. While I don't mind competition when working with more resources, working solo, I don’t need my back against the wall all the time.
✅ Goal 3: Boilerplate
Initially, I planned to use much of Canopy's codebase for non-proprietary elements and create a standard boilerplate things like sign-up, subscriptions, and admin (which would let me contribute back improvements).
However, given my time constraints, I'm reluctant to spend excessive time cleaning things like our complex billing system that covers B2B, B2C, and enterprise solutions.
I've decided to go with Jumpstart Pro. There are several things I plan to remove/change, but for now, it's enough. I’ve also wanted to learn Tailwind for a while, so it was nice that Jumpstart Pro came with Tailwind already.
I’m also sticking to Rails (The One Person Framework ™️).
I’ve been tempted to make the jump to Python because of how readily available tooling is for their ecosystem, but the fastest way to ship nothing is to pick a bunch of things you are not expert on. It’s too easy to underestimate all the ecosystem details that you need to know in order to ship a product that is stable enough to require nearly no-maintenance.
In this case, my goal is to focus on the behavior of LLMs (different models, prompt architecture, cost, reliability, etc.) and not mess around with different frontend or backend technologies.
✅ Goal 4: Canopy - Monitor the Alpha of AI Assistant
We launched our AI Assistant in Alpha last week. So, I focused on getting visibility for how the system is performing. We haven't fully committed to an evaluation strategy yet. For now, we managed to see how satisfied people are, along with some basic analytics and reporting on content that we didn't have for RAG.
For “eval” this looks very promising: https://athina.ai
I’ve seen their product at a meet-up in SF last week and found it a better fit for our needs than other tools I'd considered (open source solutions like OpenAI’s or Ragas.)
❌ Goal 5: Ship a WIP product
I didn't succeed with this one.
Last year, I started working on a product that solves a personal use. None of the business calculations add up (cost per person vs what I can charge, TAM, competitors, etc.), but it meets my criteria of being something I want to use and something I can learn from.
I've already learned a lot from this project. It's what I worked on over the weekends last year to learn different models, prompting or RAG architecture without getting burnt out from just working at Canopy. It's ~80% done, but now I just need the last stretch to make it possible for others to sign up & buy it so I'm not paying for other people's API calls.
I'll probably need another three days to get it ready for other users. Since I chose Jumpstart Pro, I'm going to migrate the product to it and see how that goes. So, this will carry over into Week 2.
Plan for Week 2
Switching jobs, picking ideas, setting up the basics, and wrapping up existing projects is kind of like tying up loose ends. It's not the most thrilling, but it's still important.
For next week, I have 3 main goals: Launch (1) and market (2) the project I just mentioned, and to do more research on my ideas (3).