LLMs
Large language models — the core building block behind every agent, RAG pipeline and AI product I ship.
Large language models (LLMs) are neural networks trained on large text corpora to predict tokens, which makes them usable for generation, reasoning, extraction and tool use. They are the hub of my AI work: almost every other AI skill in this graph is a way of making an LLM reliable in production.
How I use it
- Akashi — a provider-neutral agent framework for Dart and Flutter that puts Google Gemini, OpenAI and Anthropic Claude models behind one contract, with failover across providers, structured output with self-repair and token usage accounting.
- NovaMX CRM — an LLM-powered lead and property classification engine running on Edge Functions on top of Supabase.
- Conversational AI Agents — multi-tenant conversational agents built during my Independent AI & Full-Stack Engineer (Freelance) work.
- Open-source labs — langgraph-langserve-starter, local-rag-chatbot and langgraph-ts-agents, each with a pluggable LLM provider and an offline fake-model mode for tests.
- Somnio Software — LLMs are part of the stack in my current role.
Credentials and courses
The IBM Generative AI Engineering Professional Certificate professional certificate covers LLM architecture, Transformers, Fine-Tuning and Prompt Engineering; the Agentic AI Engineering Specialization specialization covers building agents on top of them.