Datrix | Devpost

Inspiration

Our teammate Narmeen kept seeing the same thing everywhere she went.

She works in retail and e-commerce automation, helping companies use AI to fix their workflows. And every single time—literally every time—she'd walk into a business and see the same mess.

Data everywhere. Emails stuffed with attachments nobody opens. Excel sheets that look like they were put together by five different people who never talked to each other. Critical stuff buried in systems that don't talk to each other. It was like watching money sit on the table while everyone was too busy to pick it up.

Then she had this moment that changed everything.

She was working with this massive textile supplier—we're talking suppliers for PrettyLittleThing, Boohoo, these huge global brands. She's watching their team, and they're literally typing purchase orders line by line into their system. Invoices, logistics records, everything. By hand. For hours.

She's thinking, "There's gotta be a tool for this, right?"

Nope.

She searched everywhere. Found tools that were so rigid they'd break if you looked at them wrong. Found others that were so generic they couldn't handle the weird, specific mess that real businesses deal with every day.

That gap—between what businesses actually need and what's actually available—that's what Narmeen brought to our team. And that's why we built Datrix.

What it does

We didn't want to build another demo-perfect tool. We wanted to build something that works when Karen from accounting uploads a sideways photo of a crumpled receipt at 4:47 PM on a Friday.

Datrix has three main features that actually solve real problems:

Email Automation: A Google Apps Script hooks into Gmail and only processes emails not tagged as "Processed by script." A gatekeeper agent decides if each email is important business stuff or just spam. If it passes, the main brain figures out which database table the data belongs in and saves it there automatically.

DatrixAI Chatbot: You can feed it data through plain text or attachments, and it does the same processing magic as the email automation. But here's the difference—it tells you exactly what it's planning to do and asks for permission before updating anything. Perfect for when you want control but still want the heavy lifting done.

Data Analysis: Users pick a category of analysis, type in plain English what they're looking for, and our specialized AI agents scan their integrations, run the analysis, and create actual charts on their dashboard. We built separate agents for different analysis types because one generic agent just creates mediocre results.

The best part? Once it's running, you literally never have to think about it again. No more "Hey, did anyone process that invoice from last Tuesday?"

How we built it

Development Approach: We built 100% of the frontend and 60% of the backend in Next.js using bolt.new. Bolt.new was incredible for rapid prototyping and getting the core functionality up and running fast.

The Bolt.new Limitations: When we hit the 40% mark on backend work, we ran into bolt.new's limitations. They didn't support Supabase integration with Next.js on their platform, so we switched to Cursor for the remaining backend tasks.

Cursor for the other 40%: Used Cursor to handle Supabase integration for user auth and database management, and to finalize our AI SDK agents.

Document Processing: Created a FastAPI endpoint specifically for the heavy lifting—document parsing and OCR capabilities for those potato-quality scanned files that businesses love to send.

AI Agents: Used AI SDK with OpenAI to build our specialized agents. Each agent (gatekeeper, main brain, analysis specialists) is fine-tuned for its specific job instead of trying to do everything.

Email Automation: Google Apps Script handles the Gmail integration—lightweight, reliable, and runs automatically without any server maintenance on our end.

Challenges we ran into

Honestly? Building this was messy. Way messier than we expected.

Working full-time jobs while coding at 10 PM was brutal. But it taught us something important—we had to build this thing to actually fit into real life, not just work in perfect conditions.

The team fights were intense. Like, really intense. But every time we disagreed, we forced ourselves to dig deeper into what we were actually trying to solve. Those arguments made our vision clearer.

Documents are absolute chaos. Two invoices from the same company? Completely different formats. It was infuriating until we realized—this is exactly the problem our users face every day. So we built something that thrives on chaos instead of breaking.

Figuring out where extracted data should go in a CRM was like solving a puzzle with pieces that kept changing shape. But all those iterations taught us how businesses actually work, not how we thought they worked.

Making individual agents specialists in their tasks wasn't easy and required several iterations of the prompts. Getting each agent to focus on just their specific job without trying to do everything took a lot of prompt engineering and testing.

Accomplishments that we're proud of

We built something that actually works in the real world. Not just in demos, but with actual messy business documents that would break other tools.

*The multi-agent architecture actually works. * Having specialized agents for different tasks instead of one do-everything agent was the right call—each one is genuinely good at its specific job.

Zero-touch email automation. Once it's set up, businesses literally never have to think about data entry from emails again. It just happens.

Real business impact. We're seeing companies go from hours of manual data entry to instant processing and actually making decisions with their data instead of just moving it around.

User confirmation in DatrixAI. The fact that our chatbot shows its work and asks permission before making changes gives users confidence and control.

What we learned

Here's what hit us: Companies don't need software that works perfectly in a demo. They need software that works when real people use it in real situations with terrible data quality.

Specialization beats generalization. One AI agent trying to do everything creates mediocre results. Multiple specialized agents working together creates magic.

User trust requires transparency. People need to understand what the AI is doing with their data, especially in business contexts.

Team disagreements lead to better products. Every argument forced us to think deeper about the actual problem we were solving.

What's next for Datrix

Expanding integrations. We want to connect to every major CRM, ERP, and business tool so Datrix can fit into any workflow.

Advanced analytics. We're building more sophisticated analysis agents that can spot trends and anomalies that humans might miss and expand the analysis categories.

Mobile optimization. Because sometimes you need to process that urgent invoice while you're not at your desk.

Every business we talk to has the same problem—they're drowning in their own data. We're building the platform that gives them their time back so they can focus on actually running their business instead of just managing their data.

And honestly? We're just getting started.