Abhishek Choudhary
ai-native full-stack engineer
I build AI products on real engineering foundations — production multi-tenant SaaS, RAG pipelines, edge ML at 30 FPS on a CPU.
Stack ↘
- edge ML on a bare CPU
30 FPS
edge ML on a bare CPU
DriveAware · 3 chained OpenVINO models, async inference
- institutions, one codebase
1000+
institutions, one codebase
Simption ERP · isolated DB per tenant
- self-registering modules
20+
self-registering modules
JSON-driven plugin architecture, Laravel 13
- languages auto-fixed
6
languages auto-fixed
CodeHeal · Gemini-powered GitHub agent
The living index
This is not a diagram. It is the actual index the agent below searches — 90 chunks, 3072 dimensions, projected to two. Ask a question and watch it think.
90 chunks · 3072 dims → 2 (PCA + readability refinement) · gauges sweep to true cosine scores · notch = 0.55 — the production threshold · star placed by similarity-weighted interpolation
Fire a probe · live agent
1 embedding call · 90 chunks scored in memory · watch the map above
Try asking
Every question is embedded with the same model that built this map, scored against all 90 chunks, and answered only from what clears the threshold. You're not reading claims — you're watching the machine check them.
New arrival · Series 01
Selected work · scroll to travel
Proof, not promises.
Next step
Let's ship something real.
I'm looking for SDE roles with a high engineering bar — India or international. If your team ships real product, I want to hear about it.