class TusharKumar:
role = "AI/ML Engineer"
company = "Vansun Ventures · 91Trucks"
team = "Founder's Office (Sole AI/ML)"
location = "Gurugram, India 🇮🇳"
xp = "1+ yr · Full lifecycle ownership"
stack = [
"LangGraph", "LangChain", "LlamaIndex",
"RAG", "pgvector", "Whisper", "FastAPI",
"AWS", "Docker", "Pydantic", "MLflow"
]
building_now = [
"Enterprise MCP Copilot",
"Clinical Voice AI Pipeline",
"Financial Research Agent",
]
mantra = "Ship it. Measure it. Improve it."
╔═══════════════════════════════════════════════════════════════════╗
║ PRODUCTION AI PIPELINE ║
╠═══════════════════════════════════════════════════════════════════╣
║ ║
║ [RAW DATA]──►[NER AGENT]──►[RECONCILER]──►[DRAFTER]──►[SAFETY] ║
║ │ │ │ │ │ ║
║ Clinical Extract Validate Synthesize Judge ║
║ Notes Entities Against SOAP Note ✓/✗ ║
║ GroundTruth ║
║ ║
║ ◈ LangGraph StateGraph ◈ Pydantic Contracts ◈ HITL Gates ║
║ ◈ AgentState Memory ◈ Tool Calling ◈ RAGAS Eval ║
╚═══════════════════════════════════════════════════════════════════╝
ORCHESTRATION: LangGraph ──► Deterministic DAG, not random chat
MEMORY: AgentState TypedDict ──► Ephemeral shared context
SAFETY: LLM-as-a-Judge ──► Hallucination detection layer
OUTPUT: Pydantic-enforced JSON ──► No parsing failures
| SYSTEM | ARCHITECTURE | IMPACT | |
|---|---|---|---|
| 🤖 | Saarthi RAG Chatbot | LangChain · Sentence Transformers · pgvector | Hallucination ↓ 64% |
| 📍 | Dealer Recommendation Engine | Ball Tree KNN · PostGIS · FastAPI | Latency ↓ 82% P95 |
| 🎙️ | Call Intelligence Platform | Whisper v3 · pyannote · LLM Extraction | WER ↓ 55% · F1 0.83 |
| 📄 | Vehicle Document IDP | Unstructured.io · pdfplumber · LLM Vision | Accuracy >94% |
| 📊 | Auction ETL Intelligence | Multi-source ETL · LangChain SQL Agent | 2.5 hrs/day saved |
| 🏥 | Clinical Discharge Agent | LangGraph · Pydantic · LLM-as-a-Judge | Doc time ↓ 80% |
🔴 All systems above are live in production — zero toy demos.
〔 AGENTIC & GENERATIVE AI 〕
〔 CLASSICAL ML 〕
〔 VOICE & DOCUMENT AI 〕
〔 INFRA & DEPLOYMENT 〕
| Credential | Issuer | Link | |
|---|---|---|---|
| 🟠 | Generative AI Fundamentals | Databricks |  |
| ☁️ | Google Cloud Arcade Facilitator '24 | Google Cloud | Led 1,500+ member community |
portfolio:
- name: Enterprise MCP Copilot
stack: [MCP Servers, A2A, LangGraph, FastAPI]
status: 🔨 In Development
- name: Clinical Voice AI Pipeline
stack: [Whisper, pyannote, LLM, FHIR Export]
status: 🔨 In Development
- name: Financial Research Agent
stack: [LangGraph, RAG, LlamaIndex, pgvector]
status: 📐 Architecting┌─────────────────────────────────────────────────────────────┐
│ │
│ 🎯 AI Engineer · GenAI Engineer · Agentic AI Eng │
│ 📍 Full-time · Hybrid · Remote │
│ ✅ Roles where AI systems actually ship to production │
│ │
└─────────────────────────────────────────────────────────────┘
"I don't run experiments. I ship systems."