eCommerce · AI Delivery
A real-time food recommender, MVP'd without building a data team first.
Time to MVP~2.5 months
Context
An eCommerce startup set out to build an app that recommends food to each user based on their medical data — and to do it without first standing up an expensive in-house data-science team.
The problem
- 01Medical and personal data varied so widely across the user base that hand-coding every possible criteria combination into a recommendation was impossible.
- 02Recommendations had to be generated instantly, against live stock at the user's nearest grocery store.
What we did
- 01Integrated grocery-chain data into a near-real-time, massively parallel data store using PostgreSQL and Greenplum.
- 02Implemented secure integration and storage for sensitive medical data.
- 03Built a custom machine-learning recommender in Python, exposed via API for the client's mobile team to build against.
Results
- 01MVP launched in roughly 2.5 months — without hiring expensive data scientists and engineers up front.
- 02The recommender ran against live local inventory, delivering instant, individualized suggestions.