AWS Builder Center

I Tried to Fix India's Cold Storage Problem With AWS Open Source Tools — Here's What I Learned

Built a real-time cold storage monitoring platform for shared farmer facilities using LocalStack, the AWS SDK for Java, Cedar, and the Strands Agents SDK — all offline, all free, for a hackathon.

The stat that got stuck in my head

Everyone's heard some version of this: India loses 15-20% of its fruits and vegetables every year, after they've already been harvested. It gets thrown around a lot in agri-tech circles. What nobody really talks about is where that loss actually happens, and once I looked into it, it wasn't where I expected.
My first instinct — like most people's, I think — was "just get farmers a fridge." Turns out that's not really how it works. Individual small farmers almost never own cold storage. What exists, where it exists at all, is shared storage — run by Farmer Producer Organizations, village co-ops, government schemes like PM-KUSUM, that kind of thing. There are only about 6,700 cold storage facilities in the whole country, and they're built to serve many farmers at once, not one family's harvest.
And that's where I found the actual gap: nobody's watching these shared facilities in real time. One compressor failure, one door left open too long, and produce belonging to a dozen different families spoils overnight. No alert, no record of what happened, no way to know whose fault it was.
That's the problem I built SheetalChain to chip away at, for the First Commit hackathon (Bharat Builds Tour 2026).

I decided early not to oversell it

Before I touched any code, I made a call that ended up shaping the whole project: I wasn't going to pretend this fixes the 15-20% number. Most of that loss actually happens in transport and at markets — not in storage. If I'd tried to build something that "solves post-harvest loss," I'd have ended up with something vague that solved nothing well.
So I narrowed it down to one job: if a shared cold storage facility exists, make sure it's actually being watched. Real-time temperature tracking, alerts when something's wrong, and — because it's shared — fair visibility so one farmer isn't left wondering if their batch is even being checked.

Why I leaned on AWS's open-source tools

The constraint that shaped everything else: zero budget, everything running on my own laptop. No AWS bill, no cloud account, nothing that costs money. Honestly, that's not just a hackathon rule — it's close to reality for something aimed at FPOs and village co-ops, who aren't going to sign up for a cloud bill before they've even seen if the thing works.
This is basically why I ended up leaning so hard on AWS's open-source side of things.
LocalStack did most of the heavy lifting. It's AWS's own local cloud emulator, and it let me write real DynamoDB code — actual AWS SDK calls, same as I'd use in production — and run all of it against  localhost:4566  instead of a real AWS endpoint. Nothing about the code changes later. When it's time to actually deploy this, I just point it at a real AWS account instead.
For access control, I looked at Cedar — AWS's open-source policy language, the same one behind AWS Verified Permissions. Since the whole point of SheetalChain is that a shared facility needs fair access (a farmer sees their own batch, a manager sees everything), thinking in terms of "who's asking, what are they trying to do, what are they asking about" instead of scattering  if  checks everywhere made the whole thing click for me. I'm not going to pretend I wired up the full Cedar runtime under hackathon time pressure — I built the equivalent logic in Java directly — but the way Cedar frames the problem is what shaped how I wrote it.
The hackathon also wanted an actual monitoring agent, not just a threshold check buried in an if-statement. I used the patterns from AWS's Strands Agents SDK — the tool decorators, the way it structures an agent's capabilities — to build something that looks at temperature trends across several readings, not just one number. That distinction matters more than it sounds: a reading that spikes and comes back down in two minutes is probably someone opening a door. A reading that keeps climbing steadily is a real problem.
And I'll say this plainly, because I'd rather be upfront about it than have someone ask me later and catch me overselling: I didn't wire this up to a live LLM call. I built the trend analysis as deterministic, explainable logic, using Strands' architecture as the scaffolding around it. For a live demo where reliability matters more than novelty, that felt like the right call — and I think being honest about that is worth more than claiming "real AI" for something that's really a well-reasoned rule engine.

Where DynamoDB fits in

On the "ship it" side, DynamoDB is the actual data store — units, telemetry readings, alerts, all sitting in on-demand tables so cost tracks actual usage instead of fixed capacity I'd have to guess at. Everything about how I've built it — the schema, the queries, the SDK calls — is exactly what would run against a live AWS account. LocalStack just let me prove it all out first, without spending anything.

What I'd tell someone else building this way

If you're working on something AWS-shaped with an actual budget constraint (which, let's be honest, is most of us), don't think of LocalStack as some watered-down stand-in you're stuck using. It's a legitimate way to build and test a real AWS architecture — going to production later becomes a config change, not a rewrite.
And when you're tempted to make your AI component sound more impressive than it is — don't. I think people trust a project more when you're precise about what's actually deterministic logic versus what's genuinely learned or adaptive. SheetalChain isn't trying to be everything to everyone. It's trying to fix one real, specific gap, with tools that actually made sense given what I had to work with.


SheetalChain — built for First Commit, Bharat Builds Tour 2026.
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article