Creative Technologist building AI systems you can measure.
8 years of design craft — design systems, Figma-to-code, motion, 3D — now fused with production AI engineering. Everything I ship is measured by offline test suites: 680+ tests across 11 repositories, and every number on this page has a receipt.
Measured, not claimed.
An HVAC technician asks "What does fault code E04 mean on the X200?" — and the request flows:
AegisGate → identity, rate limits, semantic cache, model routing, cost autopilot · HVAC-Copilot → grounded answers with citations · ForensiQ → every failure named and blamed · VerdictAI → releases blocked when quality regresses · RedForge → the platform attacks itself before anyone else can.
Side services: SwarmResearch (multiagent research, hallucination rate 0.0) · Model-Distillery (big-model quality at small-model prices) · PipelineViz (the whole journey as an interactive 3D site).
Start here → AI-Portfolio — the map, the episode index, and the scorecard.
| Claim | Receipt |
|---|---|
| Gateway holds 484 rps, p95 60ms under load | AegisGate · load-profile script in CI |
| RAG answers with recall@5 = 1.0, zero safety violations | HVAC-Copilot · 38-question golden set |
| Red-team exposure cut from ASR 92% → 0% | RedForge · 25 attacks, 6 defense layers |
| Judge aligned to humans via isotonic calibration | VerdictAI · hand-computed statistical fixtures |
| Enterprise decks re-branded in 45 seconds, not 40 hours | BrandMorph · 43-test suite |
10 explainer lectures, ~6 minutes each, every keyword defined on screen — plus a GLOSSARY.md per repo. Episode index →
Python · FastAPI · pydantic · GSAP · Three.js · TypeScript · React · pytest · SQLite/WAL · Redis · Langfuse · Qdrant · OpenTelemetry
🌐 akshayjohn.xyz · ✉️ akshay3rishi@gmail.com · Bengaluru, India