danidante
Remote: Yes. Part-time freelance/contract only (10-15 h/week), evenings CET, overlapping US business hours.
Willing to relocate: No
Technologies: TypeScript, Node.js, NestJS, Python, LangChain, LangGraph, OpenAI-compatible APIs, Mistral, DeepSeek, Langfuse, Redis, Docker
Résumé/CV: https://danidante.github.io
Email: [email protected]
I help teams whose LLM features work in the demo and misbehave in production. I build and run a multi-agent voice companion for seniors (about 15 repos, DeepSeek and Mistral models, EU-hosted). Things I measured and fixed there: a classifier that went from about 6.5s to 1.26s p50 after moving it to a faster model; a parameter placement whose fix cut the reply model's output tokens by 61% on typical prompts; a gateway that accepted a cache key without applying it; an eval suite that passed 11/11 without calling the model. Fixed-price audits or scoped hourly work, 10-15 h/week, 20 years of engineering before this.
Open source: https://github.com/danidante/llm-wire-check (does your LLM endpoint actually apply the parameters you send?)