Nassim E.

SaaS B2B platform enabling car dealerships and merchants to automatically prospect private individuals selling their vehicles on major French classified ad sites.

The robot contacts sellers based on precise criteria and generates inbound calls effortlessly.

I designed the technical architecture from scratch: a decoupled monorepo separating the Next.js app from a standalone API, with a queue and retry system via BullMQ and Redis to robustly manage background jobs.

The core of the product relies on a scraping engine coupled with a precise targeting system – vehicle type, geography (PostGIS) – and complete automation of prospecting sequences.

I also managed a team of 2 developers throughout the project.

Platform connecting freelance/job-seeking actuaries with insurance companies, in a market previously lacking a dedicated tool.

I designed the technical architecture from scratch on a full-stack Next.js basis, leveraging Server Actions and Route Handlers for a coherent codebase without over-engineering. The core of the product is based on a scoring algorithm ensuring the matching between actuary profiles and company offers.

I integrated Stripe for monetization, Resend for transactional emails, and Supabase as the data layer with Drizzle as the ORM.

I also managed a team of developers throughout the project.

I developed a match result prediction model (win/draw vs. loss) with 5% higher accuracy than market-leading betting assistance apps.

The main added value was in feature engineering: from an initial dataset of 100+ variables, I built over 600 features including opponent strength metrics, efficiency ratios, and positional decompositions.

I used Boruta for feature selection—particularly suited for such a large variable space—and XGBoost for modeling, with cloud-based MLflow to automate model selection and parallelize experiments.