Case Studies

Long-term engagements. I maintain a selective client roster — each gets a dedicated senior engineer, not a project handoff.

Drone Technology · Azerbaijan

Olimp UAV

Drone soil erosion segmentation — 89% mIoU, 79% erosion IoU, on-prem.

Context: Azerbaijan's first professional drone company, operating survey UAVs for industrial corridor patrol.
Problem: Manual review of UAV flight imagery for soil erosion — slow, subjective, impossible to scale across 443km.
Built: Custom CVAT-annotated dataset; SegFormer-B2 and DINOv2+UPerNet; on-prem post-flight inference pipeline with geolocated PDF reports.
Outcome: 89% mIoU / 79% erosion IoU. Automated geolocated erosion reporting after every flight.

Computer Vision Object Detection Segmentation

B2B Software · Belgium

Enfocus

Multi-year AI function — churn, forecasting, GenAI, and agentic automation.

Context: B2B software company (Belgium). Engagement since 2020, originated in industrial PhD. Finance, Product, and Marketing teams all depend on AI-built systems.
Problem: Churn risk invisible until too late. Revenue forecast built in spreadsheets. No AI product access across teams.
Built: Churn prediction on Apache Spark (full population); revenue forecasting pipeline (Salesforce + Zuora → PowerBI); multi-product GenAI assistant on Azure (OpenAI, AI Search, Cosmos DB); agentic automation; MCP server integration; CI/CD via GitHub Actions; App Insights monitoring.
Outcome: Finance, Product, and Marketing run on systems I built. Engagement renewed multi-year.

Predictive ML Big Data Generative AI Agentic Systems

Maritime Intelligence · Belgium

Tethys Naval

Maritime port intelligence SaaS — architecture to production.

Context: Belgian maritime tech startup. Fragmented port risk data across dozens of sources — incidents, weather, vessel patterns.
Problem: No unified AI-powered view of port risk for maritime planners.
Built: Multi-source LLM orchestration; AI port scoring across 5 risk dimensions; FastAPI + asyncpg + PostgreSQL/PostGIS + Redis; JWT auth; Docker + AWS App Runner; automated geolocated PDF reports; Sentry monitoring.
Outcome: Live SaaS platform, architecture to production.

SaaS Architecture AI Integration Full Stack

Academic

ISG Lille

Big Data Analytics course — Master's level, taught in French. 60+ hours delivered to date. Topics: Apache Spark, distributed processing, analytics at scale. Teaching alongside consulting, not instead of it.

Not sure which engagement fits?

A 30-minute call is usually enough to figure out the right shape. No commitment required.