Software AtlasAI Engineering
How language models behave, how to give them the right context with embeddings and RAG, how agents call tools, and how to test, ship and secure LLM features. Every demo runs on precomputed data, so nothing calls a live model.
How LLMs Work
Tokens, context windows, attention at an intuitive level, temperature, and why they hallucinate.
Embeddings
Meaning as vectors, and cosine similarity.
Vector Databases
HNSW, recall vs speed, pgvector vs dedicated databases.
Retrieval-Augmented Generation
Chunking, hybrid search, reranking, citations — and how RAG fails.
Prompting
System prompts, few-shot examples, and structured output with JSON schemas.
Agents & Tool Calling
Tool calling, the agent loop, MCP, and when an agent is overkill.
Evaluating LLM Apps
Testing non-deterministic output, golden sets, catching regressions.
LLMs in Production
Streaming, caching, retries, cost and latency.
AI Security
Prompt injection, data leaks and guardrails.