MarginMind

Inspiration

In meetings someone pulls up a spreadsheet, everyone argues about what it means, and the meeting ends with "let's revisit next week." Meanwhile, margin is leaking and no one is acting

What it does

MarginMind is like having a sharp commercial analyst available 24/7 - one who never needs context and always shows their work. You upload your sales, pricing, and cost data. Within seconds, you get a ranked list of your top 10 highest-impact actions for the week things like "increase price on this product line," or "prioritize upselling this customer segment." Every recommendation comes with an expected revenue or margin impact, and if you click on it, MarginMind explains why it's suggesting that in plain English

How we built it

FastAPI handles the backend. PostgreSQL for the database. Prophet does the demand and revenue forecasting and a custom scoring engine that ranks every possible commercial action by expected margin impact, adjusted for confidence and urgency. Score=ΔCM×Confidence×Urgency For the AI explanations, we used a self-hosted LLM via Ollama. The model explains from computed metrics, not from general knowledge. The frontend is React with Recharts, and the whole thing is containerize and runs in Docker Compose

Challenges we ran into

The biggest challenge was dirty data. Real CSV exports from real businesses are a mess with wrong date formats, missing columns, inconsistent product names, weird currency symbols. Building a validator that caught all of that gracefully (without just throwing an error and giving up).

What we learned

The best AI products are the ones that make humans feel smarter, not replaced. MarginMind doesn't make the decision for you. It gives you the confidence to make it yourself.

What's next for MarginMind

The real vision is direct integrations plug into your CRM, your ERP, your e-commerce platform so MarginMind is always working in the background. And eventually, outcome tracking so MarginMind learns whether its recommendations actually worked and gets sharper over time.

Built With

• axios
• docker
• dockercompose
• fastapi
• jwt
• llama
• nginx
• numpy
• ollama
• pandas
• papaparse
• postgresql
• prophet
• pydantic
• python
• react
• recharts
• restapi
• scikit-learn
• sql
• sqlalchemy
• statsforecast
• tailwindcss
• typescript
• vite