FEBINAHMED
PRODUCTION AI — DISCOVERY → DEPLOYMENT
I build AI products — and ship the systems behind them.
Founder & ML engineer · Bangalore · Shipped, not slideware.
What I ship.
Real products with real customers, owned end to end — discovery, build, deploy, harden. The proof isn't a deck. It's running right now.
RESTAURANT DM
Design a menu wall in the browser; watch it change on the restaurant's screens in real time. Kiosks stay online through network drops, and the whole platform is self-hosted behind Cloudflare tunnels.
STACK — FASTAPI · POSTGRES · REDIS · REACT ×3 · RASPBERRY PI
EDIT IN BROWSER → WALLS UPDATE LIVE · KIOSKS SURVIVE NETWORK DROPS
CATCHES DEFECTS ON THE PRESS, IN REAL TIME — NO CLOUD ROUND-TRIP
PRINTSIGHT
Computer-vision quality control for textile printing presses — catching ink-coverage defects on-device, in real time, on a Jetson. Built from direct customer discovery in Tiruppur's print cluster, and designed to run under a small monthly budget.
STACK — JETSON · PYTORCH · OPENCV · EDGE INFERENCE
A vague brief in. Production out.
Every capability below is backed by something real that shipped — no classroom exercises, no toy repos.
Customer discovery on a Tiruppur press floor → PRD, ROI model, edge deployment. Research → live production for Restaurant DM.
FastAPI · Postgres · Redis · three React frontends · Cloudflare tunnels · CI with gitleaks and one-task-per-commit discipline.
Single-node self-hosting behind tunnels, offline-resilient Pi kiosks, and a full homelab — Home Assistant, monitoring, alerts.
A custom Indian-accent Whisper LoRA with local TTS; multi-modal RAG on local GPUs — LLaMA-Factory, Ollama.
Multi-document RAG with graph generation on local GPUs; document extraction and segmentation feeding retrieval at scale.
Multi-agent dev workflows — planning and execution agents, custom tooling, MCP connectors; CrewAI in anger, daily.
Ink-coverage defect detection on Jetson; Raspberry Pi kiosks that keep serving when the network doesn't.
The depth behind the products.
Four years of Document AI R&D — systems that ran in production long before the founder title.
Vision, language and layout signals fused into one production classifier for Document AI — holding 95% accuracy on live traffic.
Segmentation pipelines that turn raw scans into structured, queryable data at 90% — the unglamorous work that makes everything downstream possible.
Custom OCR training runs, evaluated and iterated until error rates dropped by a quarter on production documents.
Multi-document, multi-modal retrieval with graph generation — answering over enterprise archives entirely on local GPUs. No data leaves the building.
Run in production, not in notebooks.
Builds and ships.
Building agentic workflows in production.
Multi-tenant menu-board SaaS — live at Paragon Restaurants, Kerala. PrintSight in customer-validated development.
Led R&D for Document AI: hybrid classifiers, segmentation pipelines, OCR training, multi-modal RAG on local GPUs.
Amrita Vishwa Vidyapeetham.
Ahmed, F. & Radhika, G. (2023). Snake Intrusion Detection System Using YOLOR. Inventive Communication and Computational Technologies (ICICCT).
DOI 10.1007/978-981-19-4960-9_32 ↗
I build AI systems and the products around them. My flagship, Restaurant DM, is a multi-tenant digital menu-board platform running live in Kerala — a FastAPI backend, three React frontends, and Raspberry Pi kiosks that stay online through network drops, all self-hosted behind Cloudflare tunnels.
Before founding it I spent ~4 years shipping production ML: at Ninestars I led R&D for Document AI — a hybrid CV+NLP+layout classifier at 95% accuracy, segmentation pipelines that turn scans into structured data at 90%, OCR trained until errors dropped a quarter, and a multi-modal RAG chatbot on local GPUs. I own systems end to end: discovery, prototype, deploy, harden.
Today I'm at Skillsoft building agentic workflows, and building my own products on the side — edge CV, local speech systems, a self-hosted homelab. M.Tech in AI (Amrita, 2022). Bangalore.
OPEN TO CONVERSATIONS ↓
THE ASK
Open to conversations.
Products, roles, or a hard problem that should already be in production — say hello.