I BUILDAI PRODUCTSFROM SCRATCH.
Ideation → AI → backend → infra → SEO → ship. End to end, often solo.
A Generalist
Who Ships.
I'm most at home taking a blank repo to a live product. In nearly every role I've been the sole founding engineer — I built ProfSam, my family's ed-tech company, entirely solo, and shipped 12+ production GenAI systems. I live deep in Generative AI and stay fluent across AWS & GCP. What I like most is the whole arc: ideate → build → ship → measure.
PROFSAM — an ed-tech company, built from scratch.
My father's education company. I designed and built the entire thing — marketing site & SEO, a CBSE learning platform, an AI admissions chatbot, an AI video studio, the data pipelines, and all the cloud infrastructure underneath. One person, every layer, all live.
Next.js + Sanity CMS, dynamic sitemaps, JSON-LD, TNEA cutoff tools.
CBSE Class-12: 1,500+ MCQs, multilingual video, notes, flashcards, progress.
Natural-language → SQL over 1,000+ colleges & 150k+ seats.
Vertex AI Veo 3.1, keyless AWS↔GCP auth, Fargate rendering.
Merges 4 sources into Postgres + OCR (YOLOv8 + PaddleOCR).
Sheets-driven pipeline → S3/YouTube, scrapers, OCR.
Selected Work.
NL-to-SQL Facility Intelligence Platform
Natural-language → SQL over 20+ facility schemas; schema retrieval → few-shot prompt → Claude → Postgres, with semantic caching & model routing.
Agentic Walkthrough Assistant for Enterprise IWMS
An agent that autonomously learns a customer's asset-management system; browser automation drives the live UI while Claude maps it into a Neo4j + Pinecone graph; code-enforced never-mutate safety.
Multi-Tenant White-Label AI Chatbot
WebSocket → router → per-tenant vector KB → Claude synthesis; embeddable widget, self-serve onboarding, tenant isolation & PII redaction.
Streaming Multilingual Voice Agent
Audio → Chirp STT (auto language detect) → Gemini Flash → Journey TTS, streamed; generation starts on interim transcripts.
THE ~60% STORY.
Drag the controls — watch the cost drop, the way semantic caching + model routing did it for real.
Real design: Redis semantic cache on sentence-transformer embeddings + Claude Haiku for lookups, Sonnet for complex analytical SQL with dry-run validation.
WATCH THE AGENT WORK.
An agent that learns an enterprise IWMS by driving its live UI — safely.
⛔ Code-enforced (not prompt-based) never-mutate guard, verified live in production. Cut per-form extraction ~2.7× (≈20→7 min).
The Whole Stack.
AI & Automation Engineer
Sole founding engineer of the flagship platform; scaled engineering by recruiting & leading a 5-dev team.
Senior GenAI/ML Engineer → AI/ML Engineer
Led the GenAI practice (7 engineers); 12+ production systems; 17% avg efficiency gain.
Generative AI Intern
Scaling my flagship platform to its next release while leading a 5-dev team, growing ProfSam & ScoreLab, and going deeper on agentic systems. Open to ambitious 0→1 problems.
BUILT TO ITS OWN STANDARD.
- ✓SSG / ISR pre-render
- ✓JSON-LD (Person + Article)
- ✓sitemap.xml + RSS
- ✓dynamic OG images
- ✓semantic HTML + WCAG AA
- ✓Core Web Vitals green
Let's Build
Something.
Hiring, collaborating, or curious how I'd approach your problem? I reply to everything.