A personal roadmap of ideas, projects, learning goals, and experiments.
Prototype a retrieval pipeline that runs fully on-device using a small embedding model, no external API calls. Mostly about understanding the trade-offs of **local** inference versus hosted APIs.
RAG · LLM
Spend a weekend evaluating whether a small agent loop is a better fit than a plain cron job for a task that needs judgment, not just a trigger.
Agents · Experiments
A small structural evaluation suite that runs automatically before any prompt change ships — checking length, structure, and subject coverage rather than full semantic correctness.
Evaluation · Infrastructure
A companion piece to the evaluation post — see [Notes on Evaluating LLM Output Quality](/writing/notes-on-evaluating-llm-output-quality) — this time focused on retrieval rather than generation.
Writing · RAG
A collection of small, self-contained AI experiments that don't need to become full projects — just a place to try an idea and write down what happened.
AI · Portfolio
Go past the client library and actually understand how approximate nearest-neighbor indexes (`HNSW`, IVF) trade off recall against latency.
Learning · Infrastructure
Pull shared chunking and embedding logic into a single internal package instead of copy-pasting it between experiments.
Infrastructure · Tooling
Replace the old cyberpunk-terminal portfolio with a calmer, typography-first design in the spirit of Linear, Vercel, and Stripe Press.
Portfolio · Design
Warm off-white and lavender in light mode, charcoal and the same lavender accent in dark mode — one consistent token set across every page.
Portfolio · Tooling