Essays on AI agents, full-stack systems, and building in public.
A deep dive into how I built an open-source Claude Code plugin that makes AI agents write code exactly the way I would - silently, consistently, across every project and every session.
Why I'm building Doxa, an AI medical memory layer for India's 800,000 paper-based clinics, where doctors see 40 patients a day and remember none of them after they walk out.
I stitched together 130+ skills from Anthropic, Vercel, GStack, and Matt Pocock into a single Claude Code environment that auto-commits itself, here's why it matters.
A year onsite at a Bangalore startup teaches you a few things. These are mine, and what I've been building since I left.
Transitioning from simple performance metrics to a robust Reliability Science framework for evaluating long-horizon LLM agents.
Context Engineering is the art of curating the optimal set of tokens to maximize an agent's attention budget.
Overcoming the Long-Running Agent Problem with a Two-Fold System and structured state management.
Spec-Driven Development turns documentation into executable blueprints for AI agents.
How NRQL predictions use Holt-Winters exponential smoothing to forecast time series metrics.
From $1.00 to $0.01 per slide — redesigning an AI presentation pipeline with structured outputs and prompt chaining.
Architecture decisions behind a real-time chess platform with sub-50ms move propagation and ELO matchmaking.