Case study · Data engineering & AI · 2025–now

Procurement intelligence with RAG & machine learning

Public procurement data is open but unusable in raw form: millions of records, inconsistent formats and no view of who competes with whom. I built the system that turns it into competitor reports, dashboards and pricing signals for a data consultancy's clients.

Context: Data consultancy, remote from Geneva
Clients: Manufacturing & security sector (confidential)
Scale: 1M+ contracts · 8k+ competitors

What I built, end to end

Simplified architecture. Client names, proprietary data and code are not shown.

Data engineering. Web scraping, API integration and ETL pipelines that structure federal procurement records (such as FPDS and DIBBS) into a relational model I designed from ERDs to user flows.
Knowledge base. A RAG and graph knowledge base over the structured data, used for business intelligence and automated competitor reporting.
Analytics. BI dashboards in Preset on SQLite to follow procurement trends for manufacturing and public-sector clients.
Decision support. Machine-learning and statistical models for pricing and bidding recommendations inside a custom CRM, plus Apollo API enrichment for prospecting workflows.

Why it matters

The same data powers three audiences: analysts who need the full picture, executives who need a dashboard, and sales teams who need a recommendation. Designing the data model once, then serving it through retrieval, BI and ML, keeps every answer consistent with the same source of truth.

Related public work: corruption-risk analytics on Rail Baltica procurement, a CEU project applying red-flag indicators to EU tender data.