I build AI and data systems that pay for themselves — for operations-heavy businesses where messy real-world data is costing money every day.

Founder of Deep Network Solutions Ltd · Co-founder of Finnlytics.

• Loco Reuse (System Loco, Lead DS) — LLM + geolocation pipeline that found end-of-life IoT devices in the field, identified the holding business, and made recovery routine. £1.2M/year saved, on a self-hosted Llama model with £0 spent on LLM API tokens.
• Battery health prediction — DBSCAN + SageMaker. 40% reduction in support tickets.
• MongoDB → Redshift migration — analytical throughput up 10×.
• Jira automation — 70% manual triage time removed.
• Geo-classification — flagged high-risk supply-chain dwell zones before incidents.

Python · PyTorch · scikit-learn · AWS (SageMaker, Bedrock, Redshift, Glue, Lambda) · Postgres · MongoDB · Docker · OpenAI / Anthropic / self-hosted Llama (RAG, evals, prompt + context engineering) · YOLOv8 · TypeScript / Next.js when needed.

• data-onboarding-and-reporting — production-style data onboarding & quality reporting pipeline. Python · FastAPI · pandas · WeasyPrint. Tests + CI in Docker.
• More repos landing soon — call-assistant backend (FastAPI + Twilio + OpenAI) and a small evals harness.

I run paid two-week AI Opportunity Audits (£2,500 fixed) that map your data, rank the highest-leverage AI/ML opportunities by £-impact, and end with a clear "build this next" recommendation. From there we either ship the build together or you take the roadmap and run.

Currently taking on 1–2 new engagements. UK and remote globally.

🌐 kouroshamouzgar.com · 📩 amouzgarkourosh@gmail.com · 🔗 LinkedIn · 📅 30-min call