4 autonomous AI agent pipelines in production. 95-agent fleet. 3,147 businesses processed. Zero downtime.
🔭 I deploy multi-agent AI systems into real business environments — not prototypes, not demos. Production pipelines running 24/7 with self-healing monitors, dead-letter queues, and model routing.
💬 Ask me about agent orchestration, multi-agent architecture, self-healing pipelines, GTM engineering, or turning a single spec into a deployed SaaS product.
📫 Open to remote: Forward Deployed Engineer · Agent Orchestrator · GTM Engineer · AI PM roles worldwide. DM me for a live demo.
⚡ I went from mechanical engineering to orchestrating 95 AI agents. Portfolio > résumé — here's what's live.
| Pipeline | What It Does | Scale | Stack |
|---|---|---|---|
| Ad Scout | Autonomous lead gen — discovers Meta ad spenders → scores → enriches → sends gap-matched outreach | 1,570 leads · 243 enriched · 5-stage pipeline | Python · Supabase · DeepSeek · Resend |
| TradeFlow | B2B trade orchestration — plumbing/HVAC businesses scored across 5 gap dimensions → 11 offer types | 3,147 businesses processed | Python · Outscraper · SQLite · Twilio |
| Sayso | Voice-to-work-order AI copilot — field technician speaks → Whisper transcribes → GPT structures → ServiceTitan | 97% accuracy · 99 Lighthouse score | Next.js 16 · OpenAI · Supabase · Vercel |
| Foundry | SaaS factory — spec-to-deployment pipeline using parallel agent workers + PLAN.md methodology | 4 products built | Expo RN · DeepSeek · Supabase |
Architecture highlights:
- Self-healing monitors — auto-detect agent failures, dead-letter queues prevent data loss, auto-retry with model fallback
- 95% cost reduction — SQLite local mirror architecture eliminated $500+/month Supabase egress
- Mission Control dashboard — real-time fleet health at pipeline.sujitbuilds.com
| Extension | Platform | What It Does | Status |
|---|---|---|---|
| Retinue CRM | Shopify App Store | AI-powered CRM for Shopify merchants — 499 verified leads, 38 qualified (≥70% Shopify match), lead source intelligence preserved. PocketBase + Telegram bot. | 🟢 Live |
| RecoverOS | Chrome Web Store | AI posture correction Chrome extension — on-device computer vision, 48 correction features, Stripe monetization. Zero data leaves the device. | 🟢 Live |
- BankGaadi.com — Vehicle repossession auction aggregator. FastAPI + Next.js + Playwright scrapers.
- InstantPlan — AI floor plan generator from natural language prompts. SaaS for homeowners, builders, real estate.
- DrAdityaMD — Medical practice website with auto-blog via DeepSeek API.
- LullaLog Baby — Voice-first, privacy-focused baby tracking app for iOS.
I don't just build AI products — I build the systems that sell them.
| Capability | Tools | Impact |
|---|---|---|
| Pipeline automation | Python · n8n · Clay · Zapier | 0 manual intervention in lead workflows |
| Revenue infrastructure | Supabase · Resend · Twilio | Multi-channel outreach: email + voice + SMS |
| Lead scoring & enrichment | Custom scoring engine · Outscraper | 5-dimension gap analysis across 3,147 businesses |
| Cost engineering | SQLite mirroring · Model routing | 95% infrastructure cost reduction at scale |
🎓 AI/ML Engineering Portfolio — 20 Projects (click to expand)
Analytics Vidhya GenAI Pinnacle Plus: Classical ML → Deep Learning → NLP → LLM Fine-tuning → RAG → AI Agents → Multi-Agent Systems
| Project | Stack | What I Built |
|---|---|---|
| Health Insurance Classification | scikit-learn, XGBoost | 4 classifiers + feature engineering + clinical data cleaning |
| Foundational ML × 4 | scikit-learn, XGBoost | Healthcare + Employee turnover + NYC Taxi 729K rows regression |
| Water Quality DL | TensorFlow, Keras | Dual-task DNN (regression + classification) |
| Project | Stack | What I Built |
|---|---|---|
| NLP Sentiment: TF-IDF → BERT | PyTorch, HuggingFace, BERT | RNN/LSTM/GRU/BERT comparison on 1.6M tweets |
| QLoRA BERT Fine-Tuning | PEFT, BitsAndBytes | 4-bit NF4 quantization + LoRA — trains ~0.1% of params |
| BPE Tokenizer from Scratch | HuggingFace tokenizers | 30K vocab BPE on WikiText-2 |
| Project | Stack | What I Built |
|---|---|---|
| RAG from Scratch | FAISS, MiniLM, Flan-T5 | PDF → MMR retrieval → local LLM + source attribution |
| Advanced RAG — LlamaIndex | LlamaIndex, LlamaParse, Cohere | 3 embeddings × 2 LLMs + SubQuestion + Cohere reranker |
| Project | Stack | What I Built |
|---|---|---|
| Financial Report Analysis | Claude Vision | 5 techniques on real financials: zero-shot, CoT, role-based, JSON, multi-image |
| GenAI Universe Report | Python, RAG pipeline | 9-section industry report ($5.1B→$47B market) — auto-generated .docx |
| Project | Stack | What I Built |
|---|---|---|
| ReAct Agent from Scratch | Groq, Tavily | Custom Reason→Act→Report loop — zero frameworks |
| Travel Assistant | LangChain, GPT-4o-mini | AgentExecutor + live weather + DuckDuckGo search |
| Study Assistant Quiz Generator | LangChain SequentialChain | PDF → Summary → MCQ quiz pipeline |
| Yahoo Finance API | Flask, yfinance | REST API: live prices, OHLCV, Sharpe/drawdown/MA analytics |
| Project | Stack | What I Built |
|---|---|---|
| Advanced LangGraph Agents | LangGraph, Claude Haiku, Groq | Math tool-use loop + multi-agent router |
| Competitor Analysis Agent | LangGraph, Claude Sonnet, MemorySaver | Persistent memory agent for retail competitive intelligence |
| AutoGen Agents × 4 | AutoGen 0.4, Claude, GPT-4o | Reflection · FunctionTool · StateFlow · Multi-modal bill analyzer |
| CrewAI Logistics | CrewAI sequential | Logistics Analyst → Optimization Strategist |
| CrewAI Advanced | CrewAI hierarchical, CodeInterpreterTool | Learning path generator + automated code debugger |
Languages
AI / ML / LLM
Agent Frameworks
RAG & Vector Search
Frameworks
Databases & Backend
Infra & DevOps
Automation & GTM
