Inspiration
Last year, I spent hours haggling with my internet provider—only to realize I’d overpaid $30/month for 6 months. I saw friends struggle with the same: busy schedules, unclear market rates, and anxiety about negotiating. I thought: What if AI could do the heavy lifting? I wanted to build something that levels the playing field for everyday people, turning overwhelming bill jargon into actionable savings.
What it does
BillNegotiator AI lets users upload bills (PDF/images) for instant AI analysis—comparing costs to regional market averages, calculating savings potential, and generating professional negotiation messages. It tracks negotiation progress, recommends cheaper alternative providers, and visualizes total savings (minus a transparent 33% success fee only when you save).
How we built it
We started with core AI models (TensorFlow + OpenAI) for OCR and natural language generation. Frontend: React for a seamless drag-and-drop upload and responsive dashboard. Backend: Flask/Python to process bill data, integrate market benchmark APIs, and run savings calculations (e.g., $\text{Annual Savings} = (\text{Old Rate} - \text{New Rate}) \times 12$). We hosted on AWS, used PostgreSQL for temporary session data, and Chart.js for savings trend visuals.
Challenges we ran into
OCR accuracy was a huge hurdle—bills from different providers have wildly varying formats, so we had to fine-tune the OCR API with hundreds of bill samples to extract data reliably. We also struggled with regional market data consistency; we solved this by integrating multiple local market APIs and cross-verifying benchmarks.
Accomplishments that we're proud of
We built a tool that delivers AI analysis in under 3 seconds, with a 90% accuracy rate for bill data extraction. Users love the one-click copy for negotiation messages and the transparent fee structure—no hidden costs, just results. We tested it with 50 beta users, who saved an average of $780/year (aligning with our target).
What we learned
I learned how to balance technical precision with user empathy—users don’t care about OCR models; they care about easy uploads and clear savings numbers. We also realized regional market dynamics matter more than we thought: a "good" rate in California is different from Texas, so hyper-local benchmarking was key.
What's next for this project
We’re adding support for more bill types (insurance, streaming services) and integrating direct provider comparison links to simplify switching. We’ll also test a "negotiation reminder" feature (via email) to help users follow up, and explore lightweight mobile web support for on-the-go bill uploads. No login required—we want to keep it simple, accessible, and focused on real savings.
Built With
- amazon-web-services
- postgresql
- python
- react
- tensorflow
