Booting Iqbal.OS

000

  1. mounting memory engine
  2. indexing 1,059 saree embeddings
  3. arming policy gate
  4. loading 120 x 94 MFCC frames
  5. sampling logos into particles
  6. waking 24,000 particles

AI Engineer & Full-Stack Developer

I build AI that remembers, sees, and knows when to say no.

Every dot on this page is one of 24,000 particles. Scroll, and watch them rebuild my work.

Most AI demos forget you the moment you close the tab. I build systems that keep what matters, look closer than a keyword ever could, and refuse to spend a rupee they can't justify.

0backend tests passing in GUMMY OS
0saree images indexed for visual search
0randomized inputs proving SALVAGE never overspends
0speech emotion accuracy, Top 30 of 350+

AI agent · Computer vision

Saree Visual Similarity Search

Those particles are a real saree from the catalogue. Upload a photo and a LangChain agent on Gemini decides when to search, applies filters, and explains every match.

  • 1,059 images
  • 645 designs
  • 85 CI tests
  • FashionSigLIP + FAISS
Query saree
Query
Match 1 Match 2 Match 3 Match 4

Agentic AI · Razorpay AI Builder Buildathon 2026

SALVAGE, the revenue recovery agent

Nine stages in a ring, one glowing gate. When a payment fails, the LLM proposes. A pure, tested function owns every rupee, and every action is logged in a hash chain.

  1. Observe
  2. Reason
  3. Plan
  4. Policy gate
  5. Approval
  6. Execute
  7. Verify
  8. Audit
  9. Recover
  • 27 tests
  • ~20,000 property inputs
  • Exactly-once execution

Personal AI operating system

GUMMY OS

A memory-first AI system that runs on your own machine. It stores what it learns in PostgreSQL with pgvector and ranks memories by meaning, importance, recency and confidence before any of its agents reply.

Backend tests
0
API endpoints
0
Migrations
0
Frontend tests
0
Agents
0
Tools
0
GUMMY OS landing screen

Deep learning · Federated learning · Audio

Speech Emotion Recognition

That rippling grid is 120 MFCC coefficients across 94 time frames, the exact shape the model reads. Multi-head attention classifies the emotion, trained with federated learning so raw audio never leaves its client.

Illustration: each client trains locally, only weights travel.

  • Neutral
  • Calm
  • Happy
  • Sad
  • Angry
  • Fearful
  • Disgust
  • Surprised
  • 92% on RAVDESS
  • Top 30 of ~350 at GITAM
  • TensorFlow · FastAPI · Streamlit

Full-stack web

KAFA Cafe Digital Platform

The raccoon is KAFA's own logo, drawn in particles. Behind it, a mobile-first restaurant platform with a digital menu, QR ordering, table reservations and an admin dashboard.

  • Next.js 15
  • TypeScript
  • Tailwind CSS
  • Vercel
KAFA Cafe homepage

Brand web design · Motion

CHOMPY

A playful candy-brand homepage built to read like a real modern company: brand identity, illustration, motion and responsive craft. The bear you see is its hero art, rebuilt in particles.

  • Next.js
  • TypeScript
  • Motion design
CHOMPY homepage

The journey so far

That's me, rebuilt from 24,000 particles. Push through them.

Web Developer Intern

Spotmies

AWS Data Engineering Virtual Intern

EduSkills

ETL pipelines, cloud storage and data engineering fundamentals.

Data Science Virtual Intern

Cognorise InfoTech

Machine learning for sales prediction, fraud detection and business analytics.

B.Tech, Computer Science (Data Science)

GITAM University, Visakhapatnam

PythonFastAPIPostgreSQLpgvectorLangChainGeminiFAISSRAGNext.jsReactTypeScriptTailwind CSSDockerThree.jsGSAPVercel

Hiring, or need a website that works?

Open to full-time AI and full-stack roles, and freelance web projects.