I ship production AI for contact centres and warehouses. RAG, agents, voice, and the evals that prove they work.

ADPCX · Jun 2026 · Bengaluru · sole owner on the live-call agent

• RAG
• AI agents
• LLM
• Voice
• STT
• Evals
• MLOps

A live-call assistant. Speech in, grounded reply out. The person on the phone never types a search. Call forms fill from the conversation. Built as sole engineer and shipped on GPU Kubernetes in under 10 weeks.

20× speech engine · first words in 2s · one shared GPU service

• STT
• TTS
• Voice AI
• live call

Legal and ops teams ask in English. The system finds the right file, cites it, and keeps each customer’s files private. Ran on 67 legal agreements and later on 500+ document sets.

73% precision · hybrid search · rerank · citations

• RAG
• embeddings
• hybrid search
• evals

Agents run warehouse playbooks and truck booking. A person still picks the slot. 32 workflows at 22 sites. About 3 seconds on the warehouse agent.

LangGraph · tool calling · human in the loop · 4.2/5

• AI agents
• LangGraph
• HITL

Same chat, faster answers. Production LLM serving: time to first word from 5.4s to 1.5s. Chat with memory from about 12s to 2s. Streaming, not a spinner.

structured output · token caps · FastAPI

• LLM
• streaming
• latency
• Text-to-SQL 82%

What 2026 AI engineer and MLOps posts ask for. Only the terms I can walk through in an interview.

Python · Azure · Docker · Kubernetes · LangSmith

ADPCX now: voice, RAG, LLM serving. NewCold before that: agents, legal RAG, booking, Text-to-SQL. MBA Business Analytics, batch topper. Reliance intern.

Jun 2026 – now · Dec 2024 – May 2026

• shubh12pp@gmail.com
• +91 7808968464
• Resume PDF