Building MindMap Debugger for AWS First Commit — Day 1
The setup
Three days ago I landed on an idea: a contradiction-detector that reads through a set of claims and flags where they don't add up. Then First Commit — AWS's "Bharat Builds Tour", run by WeMakeDevs, gave me a reason to finally deploy something on AWS — something I'd never actually done before.
I had 4 days. I had an idea. I had zero AWS experience deploying anything real.
Here's everything that went right, wrong, and sideways on Day 1 — before I'd written a single working line of code. The first contradiction I found had nothing to do with the project.
Problem #1: Figuring out what "built on AWS" actually means for my idea
First Commit judges on something specific: "Built on AWS" is worth real points, non-negotiable. It's not enough to have a good idea sitting behind an API call — AWS has to be doing real work in the architecture, or it reads as generic.
So the first real decision of this hackathon wasn't "what do I build" — it was how do I architect this so AWS is actually load-bearing, not decorative.
Where I landed:
- MindMap Debugger (contradiction detection) — maps naturally onto structured extraction and a UI-first demo, with room to bring in AWS-native pieces like Strands and Cedar as the pipeline matures
Lesson one, before any code: know exactly what the judges are scoring, and architect toward that from the first decision — not as an afterthought.
Problem #2: I didn't have a card. AWS didn't care.
This is the part nobody warns you about.
I went to set up a real AWS account for the Ship It track — the one with live deployment on AWS. I verified on AWS Builder Center first, no problem. Then I went to create the actual AWS account for real cloud deployment, and hit this:
Turns out AWS Builder Center and an actual AWS account are two completely separate systems. I didn't know that going in. Fair enough — I went to create a real account.
Then I hit the actual wall:
AWS account signup requires a debit or credit card. I only had UPI, and that wasn't accepted at signup. And here's where I hit my first contradiction of the hackathon.
So I pivoted to Build It — the track that explicitly promises "no AWS account, no card, no bill." It's fully real, fully eligible.
Lesson two: check your hard constraints before you pick your architecture, not after.
Problem #3: If no AWS account, then no Bedrock. Then what?
Build It's tool list includes Strands Agents SDK, Cedar, PartyRock, SAM CLI + LocalStack, OpenSearch. None of them are the actual AI brain my project needed — something that reads text and reasons about whether two claims contradict each other.
I posted in the WeMakeDevs Discord asking about local model options. A community member's reply pointed me toward Groq — free tier, genuinely no card required, and fast.
Day 1's actual stack was simpler than the final architecture: just Groq, called directly. Strands Agents SDK — AWS's own, model-agnostic, so it wouldn't care that I wasn't using Bedrock — was the plan for later, once the core pipeline actually worked.
The first real code, and the first real bug
Wired up the extraction pipeline: paste text in, get structured claims and relationships out, in JSON.
First run:
A clean 404. The model name I'd used had been deprecated by Groq back in June — dead before I ever touched it. Not a mistake, just stale information colliding with a fast-moving API landscape.
Swapped in the current recommended model. Ran it again:
Real structured output. Propositions extracted, typed, ready to reason over. This was the moment the project stopped being a plan and started being a thing that actually runs.
Where Day 1 ends
By the end of today I had:
- A clear, deliberate track decision (Build It, and why)
- A working extraction pipeline calling a real model, no AWS account required
- A first real bug, diagnosed and fixed within the hour
- A strong sense of exactly which AWS-native tools (Strands, Cedar) actually belong in this architecture — and which ones (PartyRock, SAM CLI/LocalStack, OpenSearch) don't, and why forcing them in would hurt more than help
Tomorrow: the detection logic — teaching the system to catch contradictions that aren't stated explicitly, wiring the full pipeline end-to-end, and (I'm guessing, based on how Day 1 went) at least one more bug I don't know about yet.
Tech stack so far
- Python — core pipeline
- Groq API — model backend, free tier, no card
- (Coming: Strands Agents SDK to wrap the Groq call, Cedar for policy-based filtering, Flask for the UI)
Building in public for AWS First Commit. Day 2 tomorrow — the detection logic, and probably another bug I haven't met yet.
Top comments (0)