Flowmerce
AI recommends, the merchant decides.
- Duration
- 6+ months
- My role
- Product design, full-stack development & ML pipeline
- Stack & deliverables
- LightGBMscikit-learnSMOTEFastAPIREST APIMLOpsSaaS
Flowmerce is a SaaS platform that handles e-commerce return claims from end to end. A customer files a claim on a portal carrying the store's own brand; the platform cross-checks it against the merchant's return policy and the customer's history, then recommends a resolution — exchange, refund, repair or refusal — with a confidence score the merchant can accept, override, or let run automatically.
The problem
Online stores process returns by hand, one email at a time. Every claim means re-reading the return policy, digging up the order, and guessing whether the customer is acting in good faith. It is slow, it is inconsistent from one agent to the next, and it is completely blind to repeat abusers — someone refused by one shop simply moves on to the next one.
What I built
I built the whole product — the merchant back-office, the customer claim portal, the public site and its documentation, the REST API stores plug into their checkout, and the machine-learning pipeline behind every decision. A LightGBM model trained on a 50,000-row returns dataset (cleaning, feature engineering, SMOTE rebalancing, GridSearchCV tuning, then served over a FastAPI endpoint) predicts the resolution, while a separate risk score computed across every store on the network flags claims that look fraudulent. The merchant keeps the last word: auto-approval is a switch they flip once they trust the model.
Key features
- Automatic resolution recommendation — exchange, refund, repair or refusal — with a confidence score on every claim
- Cross-store fraud score: a customer refused by one shop protects all the others on the network
- Return policy configurable per product, per reason and per delay, applied automatically
- A single REST endpoint to connect an existing store, with merchant and developer documentation
- White-label claim portal running under the merchant's own brand
- Timestamped decision history and email notifications on every outcome
Screens from the product
Public site & documentation · 4
Merchant workspace · 3
Back-office & datasets · 4
Architecture & ML pipeline · 4
Mobile · 3