Senior AI Engineer & Full-Stack Developer — London / Italy / Remote EU
I built an in-house platform that trains and serves custom diffusion and language models for campaign and TV-ad assets — what used to take a production company weeks now happens in-house, at a fraction of the cost. Another project, a voice AI agent that keeps phone scammers talking to a bot instead of a victim, won a BIMA100 AI & Tech Pioneers listing, four Cannes Lions, a Shorty, a Webby and a Gold at the British Arrows.
Agencies and production companies come to me in two different moments: some have an idea worth testing before it gets a real budget, others already have something running that needs to be steadier, faster or ready to scale. Either way, the job is the same: build it properly, so it still holds up long after the pitch or the first release.
“I think in systems. Years moving between full-stack development, computer vision, machine learning and generative AI taught me to spot the same patterns wherever they turn up, and combine them in ways that aren't obvious from inside a single field.”
Four areas I keep coming back to, each one adaptable to wherever your business is right now.
Service
Custom-trained diffusion and LLM pipelines that generate on-brand assets for social campaigns and TV ads, in-house.
Service
Segmentation, detection and masking pipelines that turn raw footage and imagery into usable signal.
Service
From a fast prototype that proves the idea to a stable, scalable system your team can rely on in production.
Service
Internal tools and automations that quietly raise the ceiling on what your business can ship.
A few public projects I worked on in my own time.
Project
Python · 44★ · 12⑂A viral clips extractor that pulls the most engaging moments out of any YouTube video, then cuts, subtitles and auto-crops them to the detected faces.
Project
Python · 0★ · 0⑂A FastAPI server that serves the Z-Image diffusion model with micro-batching across multiple GPUs to maximise throughput.
Project
Go · 0★ · 0⑂A self-hosted, privacy-first cloud storage backend with semantic search and RAG - like Google Drive, but fully local on a Raspberry Pi.
Project
Python · 0★ · 0⑂An LLM trained from scratch on 100% Italian data - custom tokenizer, dense and MoE architectures, and a roadmap to fine-tune it into a proper Sicilian speaker.
Project
TypeScript · 1★ · 0⑂A modern, scalable customer service platform built with NestJS and Angular, with AI-powered interactions and multi-channel comms.
Project
Python · 0★ · 0⑂A production-ready starter template for AI services: LoRA fine-tuning, a FastAPI backend, and cloud deployment baked in.
Project
Python · 1★ · 0⑂The Python core behind an AI content platform: workers and pipelines that generate posts, articles and composed videos from defined templates.
Project
C++ · 0★ · 0⑂A local RAG system, written in C++, that indexes Wikipedia into a vector database and answers questions through a conversational chat - fully offline.
Project
Python · 0★ · 0⑂A simple, from-scratch implementation of a diffusion model in PyTorch - training loop, model definition, sampling and dataset loader.
The toolkit behind the work: full-stack development, computer vision, machine learning and generative AI.
Development
Golang · Python · TypeScript · FastAPI · Angular · Astro · Solid.js · Azure · AWS · Runpod · Docker · K8s · Vector databases
3D
Blender · RealityScan · Litchfield Studio · 3D Gaussian Splatting · 3D object generation
Computer Vision
Segmentation · Detection · Masking · OpenCV · Pillow · FFmpeg · Moviepy
Machine Learning
CNNs · Vision Transformers · Random Forest · PyTorch · TensorFlow · Reinforcement Learning · Model evaluation & benchmarking
Generative AI
Transformers · Diffusers · LLMs · Diffusion Models · RAG · LangChain · LangSmith · Agentic Development · Fine-tuning & LoRA · Model quantization & serving · Prompt engineering · RAG evaluation & observability · MCP & agent protocols
Tools
ComfyUI · GitHub Actions · Jira · Float · Adobe Suite
Most of my work is with agencies, media production companies and marketing teams — from small, fast-moving independents who need to ship quicker than a bigger studio ever could, to larger organisations.
I also work with startups and early-stage companies building AI into their product from the ground up. The scope and cost adjust based on what's needed and how much time it takes.
I have a firm view on using AI responsibly: data gets stored safely and used only with permission, no exceptions. That's a default in how I build, not something bolted on afterwards.
I'm also a strong believer in open source. I build on open models and open tooling because the community has already done excellent work worth building on, and I try to give back by maintaining and sharing open-source repositories of my own whenever I can.
Common questions from HR teams, agencies and companies.
You can reach me via email or LinkedIn.
hello.gabriele.lanzafame@gmail.com
linkedin.com/in/gabriele-lanzafame
GitHub
github.com/gfratmct
Hybrid — London, UK & Italy (by arrangement) · Remote across Europe