Hey, Iām Syed Ahmer Shah ā a Software Engineering student and full-stack developer building web apps, AI tools and sharing lessons from the coding journey.
The 'brain is a terrible database' line is a hard truth. Iām currently mid-build on Commerza (a framework-less ecommerce engine), and I learned this the expensive way.
I hit a major refactoring disaster recently where I lost 40% of my backend logic. What actually saved the project wasn't just my code backups, but the fact that I had documented my 'security primitives' (SQL transactions, auth logic, etc.) separately. It turned a total loss into a 9-hour reconstruction job instead of a 1-month failure. Iām definitely looking into the PARA folder structure for my next project to keep those architecture decisions even more accessible.
Ouch, losing 40 percent of your backend is a painful way to learn it. Glad the docs saved you, that is exactly the point, the decisions are the asset, not just the code. Good luck with Commerza.
Mine requires the smaller of the two max plans, and it's pretty similar. This doesn't require top models for everything, though. A lot of processing can be done with haiku if you write tight enough prompts, but that's not always easy when exploring new concepts.
My first version of this used opus to write the prompts and md files, and haiku to actually execute those prompts. Then had sonnet fix things along the way. I used regular Claude app for the design phase, Claude Code for the rest. I don't have the best frame of reference (I'm spoiled), but I think that would have worked with Pro usage levels.
This is similar to my own! I've recently added hooks and started logging the conversation back and fourth and I inject memories using hooks and a vector db (if I have memory mode on) into the prompt itself. I have ADHD and memory issues, so I like to be reminded of what happened in the past, but it actually seems to improve a lot of answers since so much of what I discuss with AI is related.
I did add a scripts and journal folder, both for work and both shared with other projects. I have a wiki folder, too, where all the processed files live. That's a pretty big strike against the methodology, but I think it's important that we all change the system to fit our own needs. My wiki folder is the same information found in other areas, just written from a different, technical perspective vs a lot of opinion and emotion found elsewhere. That's how it's broken up in my head, so that's how it's broken up in my second brain.
That hooks and vector db setup sounds great, especially the memory injection. And you are right, the system has to fit how your brain works, otherwise you just stop using it.
signed in after founding out about Ingress NGINX
in post with "When Projects Fail:..."
I dont know Go so I would not have been much of an use but if I would have known sooner...
Digital Strategist @ MetaLLM š
š¤ AI-driven engagement & growth.
š¬ Daily insights on the future of tech.
š Building communities, one post at a time.
Let's grow together! š
i've been using a physical notebook for morning dumps and it's wild how much it reduces that mental cache pressure. no notifications no context switching. feels like defragging my brain before the day even starts
Yes, the notebook morning dump is underrated. No notifications, no context switching, just get it out of your head. I do something similar before I open anything else.
Automation consultant. I build AI-powered workflows using Claude, n8n, and open-source tools. Sharing practical guides on AI agents, no-code automation, and cost optimization.
The "brain as terrible database" framing is a great reframe. The adjacent problem I keep running into: most second-brain systems are great at write and terrible at recall. I've had the most luck with a flat daily-notes markdown tree + grep + a small semantic search layer on top ā the structure stays dumb enough that tools (and agents) can read it, but retrieval still works because of embeddings. Heavy folder hierarchies tend to become the place notes go to die.
The 'brain is a terrible database' line is a hard truth. Iām currently mid-build on Commerza (a framework-less ecommerce engine), and I learned this the expensive way.
I hit a major refactoring disaster recently where I lost 40% of my backend logic. What actually saved the project wasn't just my code backups, but the fact that I had documented my 'security primitives' (SQL transactions, auth logic, etc.) separately. It turned a total loss into a 9-hour reconstruction job instead of a 1-month failure. Iām definitely looking into the PARA folder structure for my next project to keep those architecture decisions even more accessible.
Ouch, losing 40 percent of your backend is a painful way to learn it. Glad the docs saved you, that is exactly the point, the decisions are the asset, not just the code. Good luck with Commerza.
Where can I see all of the steps to configure the shared brain? Do I need paid subscription with Claude?
based on the look of it wou'll need a Max or at least a Pro plan
Mine requires the smaller of the two max plans, and it's pretty similar. This doesn't require top models for everything, though. A lot of processing can be done with haiku if you write tight enough prompts, but that's not always easy when exploring new concepts.
My first version of this used opus to write the prompts and md files, and haiku to actually execute those prompts. Then had sonnet fix things along the way. I used regular Claude app for the design phase, Claude Code for the rest. I don't have the best frame of reference (I'm spoiled), but I think that would have worked with Pro usage levels.
Love the Opus / Sonnet / Haiku split, that is a smart way to keep costs down. Tight prompts doing the heavy lifting is the whole game, really.
Not great but still like the tool and how it works.
Ha, fair. What did you find not great? Curious what broke for you.
Solid approach, love the mix with Obsidian and GDrive
Yes this is applicable for any usecase not just managers but also engineers / builders.
This is similar to my own! I've recently added hooks and started logging the conversation back and fourth and I inject memories using hooks and a vector db (if I have memory mode on) into the prompt itself. I have ADHD and memory issues, so I like to be reminded of what happened in the past, but it actually seems to improve a lot of answers since so much of what I discuss with AI is related.
I did add a scripts and journal folder, both for work and both shared with other projects. I have a wiki folder, too, where all the processed files live. That's a pretty big strike against the methodology, but I think it's important that we all change the system to fit our own needs. My wiki folder is the same information found in other areas, just written from a different, technical perspective vs a lot of opinion and emotion found elsewhere. That's how it's broken up in my head, so that's how it's broken up in my second brain.
That hooks and vector db setup sounds great, especially the memory injection. And you are right, the system has to fit how your brain works, otherwise you just stop using it.
I have done something similar for my personal project tracking.
Nice, what are you tracking in yours? For me the win was getting project decisions out of my head and somewhere I could actually find them again.
This is really good idea I have been thinking how to organize all my vibe coding ideas.
Yes this is applicable for any usecase not just managers but also engineers / builders.
i've been using a physical notebook for morning dumps and it's wild how much it reduces that mental cache pressure. no notifications no context switching. feels like defragging my brain before the day even starts
Yes, the notebook morning dump is underrated. No notifications, no context switching, just get it out of your head. I do something similar before I open anything else.
The "brain as terrible database" framing is a great reframe. The adjacent problem I keep running into: most second-brain systems are great at write and terrible at recall. I've had the most luck with a flat daily-notes markdown tree + grep + a small semantic search layer on top ā the structure stays dumb enough that tools (and agents) can read it, but retrieval still works because of embeddings. Heavy folder hierarchies tend to become the place notes go to die.
Totally agree, recall is where most systems fall over. Flat notes plus search beats a clever folder tree every time, in my experience.