Blog
Chanakya: A 13-Agent Portfolio Brain, Engineered for Restraint
Inside a private, solo-built multi-agent LLM trading system — an event-driven Kafka pipeline where only 9 of 13 agents ever call an LLM, a Bull/Bear/Judge debate that's really one agent wearing three hats, and the production hardening (idempotent deduplication, tiered inference routing, on-device embeddings, circuit breakers) that keeps it running sustainably, every single day.
SuperSay: On-Device Text-to-Speech That Streams in Under 200ms
How SuperSay gets sub-200ms time-to-first-audio from a model that runs strictly one inference at a time, why INT8 quantization was tried and rejected with data, the pre-launch adversarial audit that caught a real consent bypass, and the business case for running the model yourself instead of renting someone else's.
SuperZen: Engineering a Native macOS Wellness App That Doesn't Get in Your Way
A deep dive into SuperZen's architecture — a single 1Hz heartbeat driving the entire app, a profile-adaptive Focus Quality Score, privacy-first local-only telemetry, and the 36-hour hardening sprint that turned a two-day prototype into a licensed, accessible, publicly-releasable v1.1.10.
ytdld: Building a Live TUI on Top of a Library That Wasn't Built for One
yt-dlp is a synchronous, blocking library with no progress-separation, no cancellation API, and its own opinions about writing to your terminal. Here's how ytdld builds an 8-way-parallel, cancellable, retryable live dashboard on top of it anyway — without forking yt-dlp or fighting it.
BroSki: A Task Runner That Tells You Why It Reran
Why a Rust task runner needed transactional output promotion, content-addressed fingerprinting, and platform-aware sandboxing just to answer one question honestly: why did this rebuild?