Artifacts

Labs are where I turn questions into executable artifacts: notebooks, controlled experiments, small systems, and reproducible comparisons.

• LLM Post-Training Lab — controlled SFT, DPO, PPO, and GRPO experiments around reward, KL, capability retention, diversity, and preference behavior.
• Agentic Systems Lab — agent patterns, workflow spines, delegation, verification, human gates, MCP, A2A, and multi-agent orchestration.
• Transfer Learning Lab — experiments on transferability, negative transfer, forgetting, multi-task interference, modular composition, and self-supervision.
• DSPy Lab — LLM programming as specification, evaluation, and optimization; from signatures and modules to agents and observability.
• LLM Text Generation Lab — decoding, sampling, reasoning controls, self-consistency, adaptive compute, verifier loops, and search.
• Stat-ML Lab — simulation-heavy applied statistics and ML across experimentation, causal inference, Bayesian methods, forecasting, production ML, SQL/Spark, and operations research.
• Voice AI Lab — practical experiments across speech recognition, speech generation, diarization, voice cloning, streaming, and realtime voice systems.
• Frontend Lab — understanding the modern web stack by rebuilding one application from HTML through React, Next.js, React Native, auth, and shared clients.

The lab format is part of how I learn: question → experiment → learning → recorded notes.