Looking for a job

Arnab Mukherjee

@arnabmukherjee2

AI Engineer at MARS Snacking building voice and chat agents for enterprise data discovery, including a real-time voice assistant.

India

Message

At MARS Snacking, I built Marvis, a real-time voice AI for DataPulse that supports hands-free discovery and onboarding. I also built a DataPulse chatbot that searches over 5M asset records and a data-quality assistant that turns natural language into Databricks DQX rules.

At Atrina Technologies Pvt Ltd, I engineered a multi-tenant voice assistant with 0.5s response time and integrated voice agents with Twilio, Exotel, and Meta telephony. My earlier work includes meeting intelligence, finance workflow automation, compliance documentation, and computer vision and NLP projects at Bhabha Atomic Research Centre (BARC).

Experience

Work history, roles, and key accomplishments

Current

AI Consultant

Current

MARS Snacking

Apr 2026 - Present (6 months)

Built Marvis, a real-time voice AI for DataPulse on Azure Voice Live with bidirectional browser WebSocket audio and speech-to-action tools. Also built an enterprise DataPulse chatbot on LangChain deep agents and Azure OpenAI, and shipped an AI data-quality assistant for Databricks DQX rules.

Azure Voice Live WebSockets PCM16 Lang Chain Azure OpenAI Databricks DQX Unity Catalog MLFlow SQL Warehouse Vector Search Service Principal

AI Engineer

Atrina Technologies Pvt Ltd

Feb 2025 - Apr 2026 (1 year 2 months)

Engineered ultra-low-latency voice assistant achieving 0.5s response time using LiveKit with OpenAI Realtime API and Gemini Realtime API. Architected multi-tenant voice AI infrastructure and integrated enterprise telephony systems via Twilio, Exotel, and Meta platforms.

LiveKit OpenAI GPT Realtime Gemini Realtime API WebRTC VAD Twilio Exotel Meta SIP IVR Whisper Lang Chain LangGraph Qdrant GPT 4 RAG GCP Docker CI CD

Trainee – Data Science & AI

Bhabha Atomic Research Centre (BARC)

Jan 2024 - Dec 2024 (11 months)

Developed a Computer Vision-based Radiation Exposure Detection System using OpenCV and deep learning, achieving 85%+ detection accuracy. Integrated GPT-3 for automated medical report generation and engineered NLP pipelines with LangChain and HuggingFace LLMs.