Faris Elshammouty AI Engineer
I build and ship production AI systems: large language models, retrieval pipelines, and autonomous agents. I like taking an idea from rough prototype to something people can actually rely on.
I build AI that holds up in production: measured, observable, and genuinely useful to the people who rely on it.
My focus is the full lifecycle of applied machine learning: framing the problem, choosing the right model strategy, and engineering the system around it, from data pipelines to evaluation, deployment, and monitoring. I care as much about latency, cost, and reliability as I do about accuracy.
Lately I work most with large language models: designing retrieval-augmented architectures, fine-tuning for narrow domains, and building tool-using agents that do real work safely. The part I enjoy most is when a rough prototype turns into dependable infrastructure.
What I work with.
Language models
- LLMs
- Prompt engineering
- Fine-tuning
- Agents & tool-use
- Hugging Face
Retrieval & data
- RAG
- Embeddings
- Vector databases
- LangChain
Modeling
- Python
- PyTorch
- TensorFlow
- Evaluation
Deployment & MLOps
- FastAPI
- Docker
- AWS / GCP
- Model serving
Things I’ve built.
Real, shipped projects: a mix of full-stack apps and applied AI.
ChaseGuard
A full-stack banking platform built for a JPMorgan Chase and Bournemouth University program. It combines real-time fraud detection and role-based access control with financial-wellness dashboards and full audit logging, on a Django REST backend and a React front end.
Openfy
A privacy-first desktop music app with a clean, Spotify-style interface. It streams millions of tracks with synced lyrics, audio visualizers, and listening stats, while keeping your whole library and listening data on your own machine.
GALex Digital
A digital edition of a Greek-Arabic lexicon for classics researchers. It supports bidirectional search across Greek, Arabic, and transliteration, links each translation back to its source passages, and lets you cross-reference entries by root and by author.
Purbeck World Cup Hub
A World Cup 2026 prediction game for a student residence. Players forecast match results and pick a champion to climb a live house leaderboard, backed by user accounts, live score feeds, and a points engine that scores every prediction automatically.
Engineering discipline, applied to AI.
Evaluation first
I define what “good” means with measurable evals before building, so progress is real and regressions are visible.
Grounded & safe
Retrieval, citations, and guardrails keep model output anchored to sources and within safe bounds.
Cost & latency aware
The right model for the job, balancing accuracy against the budget and speed a product can actually afford.
Built to operate
Monitoring, tracing, and clean interfaces so systems stay reliable long after launch day.