David Alimi
AI Engineer at Geta.Team / Elestio
I build production AI systems - RAG pipelines, AI agents, and fullstack applications with Python and TypeScript.
At Geta.Team, I work on AI-powered products deployed on Elestio's cloud infrastructure. Before AI, I was a fullstack developer with React, Express, and PostgreSQL.
What I do
I specialize in building AI applications that actually work in production - not demos, not prototypes. Systems with verifiable sources, evaluation pipelines, and cost-aware architecture.
My current focus is Torah Study AI - a RAG pipeline on 3.5M sacred texts where hallucination is not an option. Every design decision is documented, benchmarked, and justified.
How I work
I follow a hybrid BDD + TDD workflow: write behavior scenarios first (GIVEN/WHEN/THEN), prioritize by complexity (simplest first), then implement with Red-Green-Refactor. Every feature starts with a test that fails.
For AI systems specifically, I add evaluation-driven development: relevance thresholds, LLM-as-Judge rubrics, and automated quality checks before shipping. Because on sensitive data, "it works on my machine" is not enough.
Tech stack
AI / ML
RAG pipelines (Weaviate, hybrid search, Cohere rerank), embeddings (Gemini, Qwen, local models), LLMs (Gemini, Claude, GPT), evals (Ragas, LLM-as-Judge)
Backend
Python (FastAPI), Node.js (Express), SQLite, PostgreSQL, JWT auth, SSE streaming
Frontend
React, Next.js, TypeScript, Tailwind, shadcn/ui
Infra
Docker, Elestio, llama.cpp (local LLMs), LangFuse (observability)
Why this site
I document everything I build. Not tutorials - build logs with real code, real costs, and real failures. When I estimated embedding costs at $2.60 and it turned out to be 25x more, I wrote about it. When my RAG system returned garbage answers, I debugged it in public.
No hype. No "10x your productivity" content. Technical depth with an opinion.