Honest AI Stack Matrix
A transparent breakdown of AI models, frameworks, and agent architectures used in production.
AI Stack & Expertise
A transparent view of technologies I use in production, actively master, and explore.
OpenAI (GPT-4o / O3)
Used in live customer agents with structured JSON outputs and function calling.
Anthropic Claude (3.5 Sonnet / Haiku)
Primary model for complex code reasoning, architecture analysis, and long-context processing.
Google Gemini (1.5 Pro / Flash)
Multimodal audio/video ingestion and high-throughput background processing.
Model Context Protocol (MCP)
Connecting LLM agents to internal databases, file systems, and external tools securely.
LangGraph / LangChain
Cyclic multi-agent orchestration, state persistence, and human-in-the-loop nodes.
n8n (Self-Hosted Workflow Engine)
Event-driven workflow automation connecting webhooks, databases, and AI nodes.
Hermes Agent Framework
Exploring lightweight autonomous execution environments and local tool execution.
Prompt Engineering & Evaluation
System prompt optimization, few-shot conditioning, and structured output enforcement.
Function Calling & Tool Use
Enabling agents to execute SQL queries, trigger webhooks, and call REST endpoints.
Agentic Memory Systems
Short-term conversation buffers, Redis session state, and long-term vector memory.
RAG & Vector Search (Pinecone / PgVector)
Chunking, embedding, hybrid BM25 + dense search for enterprise document retrieval.
Fine-Tuning & Local LLMs (Ollama / vLLM)
On-premise LLM hosting and task-specific LoRA fine-tuning for privacy-sensitive data.