Problem
Start with a useful capability or failure mode specific enough to evaluate.
AI Lab
Small prototypes for learning how intelligent systems behave.
A public notebook for connecting Generative AI concepts to system architecture, implementation, evaluation, and practical limits.
Lab statusLearningBuild · evaluate · revise
The method
Context
Define the data, tools, constraints, and assumptions the system receives.
System
Build the smallest architecture that can test the core idea.
Evaluation
Measure quality, groundedness, reliability, cost, and failure cases.
Limits
Record what remains uncertain and what evidence should come next.
Current entries
Prototypes and system sketches
LAB 01Planned prototype
Grounded answers with retrieval
How do chunking, retrieval, and citations affect the usefulness and traceability of an LLM answer?
RAG · retrieval quality · grounded evaluationNotebook in progressLAB 02System sketch
An agent loop that fails visibly
How should tools, state, validation, and recovery fit into a dependable agent workflow?
Tool use · state · validation · recoveryNotebook in progress