Real-Time AI Voice Agent Platform
Production conversational AI built at Zencia — streaming ASR into an LLM (Claude / OpenAI) and back out through TTS, augmented with LangChain RAG and MCP-based agentic tool-use, all served over FastAPI.
AI Engineer specialising in LLM-powered voice agents, RAG and agentic systems — with a research backbone in deep learning & computer vision. Recently shipped production Voice AI as a Gen AI Engineer at Zencia AI.
I love turning complex ideas into simple, intelligent and useful systems. My work sits where applied research meets production engineering — having shipped LLM & voice-AI products at Zencia AI, all grounded in the maths I studied through an M.Sc. in Data Science at IIIT Lucknow.
Whether it's a real-time voice agent, a physics-informed vision model or a data lakehouse, the goal is always the same: solve a real problem, end to end.
Ship things people actually use — measured, monitored, improved.
Grounded in maths, statistics and honest evaluation.
Prototype → deploy → monitor. From notebook to production API.
New domains, new tools, new questions — every single week.
Built production LLM voice agents & conversational AI — ASR → reasoning → TTS, RAG pipelines and agentic workflows (LangChain · MCP) served through FastAPI.
Statistics & Optimization — mentored 220+ students in inference, optimization algorithms and ML fundamentals.
ML, Deep Learning, NLP, Reinforcement Learning & Big Data. CGPA 7.67 / 10.
Linear algebra, probability, operations research & mathematical modelling. CGPA 8.33 / 10.
Most recent role
Zencia AI Tech Pvt. Ltd. · Lucknow
Mar 2026 — Jun 2026
Production conversational AI built at Zencia — streaming ASR into an LLM (Claude / OpenAI) and back out through TTS, augmented with LangChain RAG and MCP-based agentic tool-use, all served over FastAPI.
Burned-area segmentation from satellite imagery, with physics-informed constraints (PI-VMUNet) improving calibration over VM-UNet baselines.
U-Net / HRNet pipeline estimating Total Body Surface Area from a 6,000+ synthetic image set built with Poisson blending. Supervisor: Dr. Niharika Anand.
XGBoost / CatBoost ensemble with feature engineering, threshold tuning and MLflow tracking — deployed as a Dockerised FastAPI inference service.
Medallion architecture (Bronze → Silver → Gold) on Apache Spark & Airflow with automated ETL and data-quality validation.
Explore the full set of repositories, notebooks & experiments.
Visit @ToshitDwivedi →A stack that spans the full AI lifecycle — from LLM orchestration and voice pipelines to deep-learning research and data engineering.
Verified experience & coursework. Click any card to view the full certificate.
Zencia AI Tech Pvt. Ltd.
C-DAC Hyderabad × IIT Roorkee
Drop a certificate (PDF/PNG) in /static and it slots straight into this gallery.
Technology is not just about code and models. It's about solving real problems, improving lives, and pushing the boundaries of what's possible.
Toshit Dwivedi
Open to AI / ML / Data Science roles and interesting collaborations. The fastest way to reach me is email or LinkedIn.