Sunit Adhikary
SR. AI ARCHITECT
8 yrs experience · Jaipur, Rajasthan · 30-40 hrs/week
About
I build AI that actually ships. With 7+ years across deep learning, LLMs, agentic AI, and MLOps — I've gone from building NILM models on smart meters at Jio to architecting multi-agent RAG pipelines and leading AI product teams. I don't just prototype; I deliver production systems that handle millions of data points and real business impact.
Skills
Experience
- DATA SCIENTIST (DY. MANAGER) → FUNCTIONAL: DATA SCIENCE MANAGER | · Reliance Jio Infocomm Limited06-01-2024 – 09-01-2024
Architected the end-to-end Jio HES+ Copilot GPT-4 (finetuned via instruction tuning + CoT exemplars) for Normalizer, Query En- hancer & Router; on-prem Llama-3.1-70B (vLLM) for answer composition; coder-tuned Qwen2.5-Coder for MongoDB aggregation & plotting queries. • Agentic Execution: Multi-agent routing across billing, NILM, feeder ops, PQ, geo; JSON/function-calling guardrails; Serper API for live internet fetches (DISCOM news & alerts). • Impact: Reduced 20 siloed dashboards to a single Copilot interface; sub-2.5s p95 latency; boosted officer engagement 300% • Tech Stack: LangChain, LangGraph, Weaviate/FAISS, PyMuPDF, GPT-4, Llama-3.1, Qwen2.5-Coder, Serper API, Docker, Kubernetes, Python (pandas, numpy, matplotlib, PyTorch) • Jio HES+ Analytics Platform: Architected and deployed a large-scale DISCOM intelligence suite analysing 1 Million+ smart meter data points daily across thousands of DTs/LMVs. • Billing & Revenue KPIs: Thru rate, billing efficiency, exceptional billing, bill corrections, due-date deviation, monthly collection effi- ciency, timely collection, revenue realisation across LMV/HV categories, and segmentation of habitual late payers. • Consumption & Behaviour Analytics: Average/peak/off-peak usage (MoM/YoY), load pattern vs temperature/humidity, peer bench- marking across regions, high/low bill alerts, and COVID-era behaviour change analysis. • Reliability & PQ Analytics: SAIFI, SAIDI, CAIDI, MAIDI/MAIFI, power factor analysis & penalties, voltage scoring system, Power Quality Index (PQI), Billing Quality Index (BQI). • Event & Tamper Intelligence: Categorisation of outage, PQ, tamper, and operational events; neutral disturbance tracking; tamper analytics (magnetic, reverse current, bypass, cover open) with hotspot maps and DT-level raid candidate identification. • Asset Health & Utilisation: Detection of over-utilised DTs, stress duty cycle monitoring, poor-performing asset identification, and composite Asset Health Score with geo-maps of risky
- DEEP LEARNING ENGINEER · Dimension Nxg10-01-2018 – 04-01-2020
• Played a key role in developing state-of-the-art 3D vision systems for segmentation, classification, detection, 3D bounding boxes, and 6D pose estimation from LiDAR point clouds, deployed in the flagship product Ajnalnes. • Led R&D on motion and performance capture, creating deep learning solutions for human and hand pose estimation, deployed as a scalable web service on Google Cloud. • Advanced research on 3D avatar creation from a single RGB image, contributing to innovation in immersive and virtual environments. • Optimized and deployed computation-heavy deep learning algorithms on resource-constrained edge devices (Intel Neural Com- pute Stick, Qualcomm 845 dev kit). • Research Contributions: Co-authored 1 conference paper and contributed to 3 patents during tenure.
- TECH LEAD · Rootent Technologies09-01-2025 – 01-01-2026
Leading the National Jobs Portal for Wadhwani Foundation, architecting an agentic AI platform with RAG pipelines and multi-agent workflows for job discovery, skill-matching, and career guidance.
Education
- IIITG(Indian Institute of Information Technology Guwahati)B.TECH., COMPUTER SCIENCE AND ENGINEERING
Similar talent on Pangea
Hire Sunit through Pangea
Describe your project to the Pangea agent — see if Sunit is a fit, with transparent pricing and interviews booked straight onto your calendar. No contact details change hands until you hire.
See if Sunit is a fit