// the tools I built because the existing ones annoyed me enough to fix it myself.
// exhibit A — this very site compiling. more casts land on projects as I record them.
npm run build
drwxr-xr-x TruADHD/
live / launched Aug 2026public
An independently verified ADHD provider directory for Pune & PCMC.
WHY
Finding a qualified ADHD professional in India means weeks of guesswork against unreliable listings. TruADHD hand-verifies every provider against public credentials, with no paid rankings or ads — verification is the only gate to being listed.
IMPACT
15 verified providers across 4 categories, 6 in-depth guides, category×city SEO pages, fully prerendered and crawlable, live at truadhd.com
Persistent memory, skills, modes, and work triage for AI coding agents.
WHY
Agents kept losing the thread between sessions. Wizard became the local-first memory layer I wanted: session hooks, 20 MCP tools, 17 agent skills, working modes, PII scrubbing, FTS5 search, analytics, and optional Notion/Obsidian write-back.
A local-first Rust agentic coding CLI built around harness strength, not model faith.
WHY
Local models are weaker at tool-calling than hosted frontier models. Alchemist compensates with a bounded shared executor, jailed native tools, deterministic TOML workflows, MCP server mounting, confirmation policies, and strict crate boundaries.
IMPACT
single execution core for free-form agent loops and deterministic workflows; no unsafe/unwrap/expect/panic/todo/dbg in non-test code by workspace lint
STACK
RustLM StudioMCPTOML workflowsReedlineTokioClippy
private/internal repo
drwxr-xr-x Forge/
usable internal platformprivate
An internal ops platform that replaced four spreadsheets and a group chat.
WHY
A 3D-printing business needed order management, cost calculation, inventory tracking, slicing, CRM sync, and pricing workflows in one place instead of duct-taped spreadsheets, so I scoped and built the operating system around it.
A YouTube automation platform where every content action is a tracked task.
WHY
Content workflows get messy once transcripts, ideation, generation, analytics, moderation, and publishing all become separate moving parts. CtrlAlchemy treats the whole thing as observable tasks with specialized agents, evals, tracing, and local/prod model routing.
IMPACT
15+ specialized agents, task-first architecture, OpenTelemetry/Prometheus observability, A2A signing, Vertex AI + Ollama model paths
A YouTube channel for the engineering layer nobody puts in the demo.
WHY
The interesting part of senior engineering rarely makes it into tutorials: the tradeoffs, the dead ends, the "why we didn't do it that way." So I started filming that part instead.