Open to opportunities

Bhuwan Jain

AI Engineer & Full Stack Developer

I build LLM applications, agentic systems, and production ML — from data pipelines to deployment. Currently shipping a multi-tenant Voice AI SaaS at Chittanshai.

Get in touch Resume

About

I'm an AI engineer and full-stack developer focused on LLM applications, agentic systems, and computer vision. I've shipped production work spanning a deepfake detection model with 88.4% test accuracy, a multi-agent autonomous AI platform, and a live multi-tenant Voice AI SaaS.

Comfortable across the full ML lifecycle — data pipelines → modeling → deployment → monitoring — with an emphasis on scalable, reliable systems.

Integrated MSc in Mathematics and Computing from BIT Mesra, Ranchi (2020–2025).

Experience

Feb 2026 — Present
Junior AI Developer · Chittanshai
- —Designed and deployed a production-grade Voice AI SaaS platform enabling businesses to launch AI voice agents for customer service; integrated the Bolna telephony API with a FastAPI backend and Next.js frontend.
- —Engineered prompt engineering pipelines for LLM-powered conversational flows, tightening response consistency and reducing fallback handoffs across multiple customer-service use cases.
- —Built an Oil India PQ Drafting Agent — a production RAG agent that drafts replies to Parliament Questions on Oil India's E&P operations over a 50+ document ministry corpus, using hybrid retrieval (Qdrant dense + BM25 + RRF fusion) with FlashRank reranking and a LangGraph drafter with citation validation.
- —Currently building an AI-powered stock market insights platform that delivers real-time market analysis and lets users place orders through a conversational interface.
May 2024 — Aug 2024
Summer Intern · Pune Knowledge Cluster (PKC)
- —Built a MERN-stack quiz platform that crowdsourced 5,000+ galaxy image classifications for downstream ML training; designed the MongoDB schema and built the SQL Server → MongoDB ingestion pipeline.
- —Implemented JWT-based authentication, an admin dashboard, and a randomization algorithm ensuring each image appeared in no more than 100 unique quizzes.
- —Cut image retrieval latency by 40% via Redis caching and containerized the full stack with Docker for one-command deployment.

Projects

Deepfake Detection

Master's Thesis

ResNeXt-50 + LSTM hybrid combining frame-level feature extraction with temporal sequence modeling for deepfake detection on FaceForensics++ and DFDC datasets. Reached 88.4% test accuracy, outperforming pure-CNN baselines on temporally manipulated clips. Containerized and deployed on AWS EC2 behind a Flask API with a React/TypeScript UI for real-time video analysis.

PyTorch
OpenCV
AWS EC2
Flask
React
TypeScript

Autonomous Agentic AI System

Data-Driven Decision Making

Multi-agent system (Planner–Executor–Critic) that decomposes complex goals, selects tools dynamically, executes sub-tasks, and self-validates through long-term memory and feedback loops. Built a grounded RAG pipeline with LlamaIndex + Qdrant for source-anchored answers, shipped as Dockerized FastAPI microservices.

Python
LangChain
LlamaIndex
HuggingFace
Qdrant
FastAPI
Docker

Voice AI Platform

Multi-tenant SaaS

Architected the multi-tenant SaaS layer for the Voice AI product — workspace isolation, JWT-based auth, webhook event handling, and a third-party API abstraction layer that decouples the platform from any single voice/telephony provider, allowing the underlying engine to be swapped without client-side changes.

FastAPI
PostgreSQL
Next.js
TypeScript
LiveKit
JWT

Skills

Languages

C++
Python
TypeScript
JavaScript

AI / ML

PyTorch
Scikit-Learn
Pandas
NumPy
NLTK
HuggingFace
LangChain
LangGraph
LangSmith
Langfuse
LlamaIndex
RAG
OpenCV
Google GenAI
LiveKit

Tools & Infra

Git
Docker
Kubernetes
FastAPI
Flask
Postman
Qdrant
Ollama
Kafka
AWS

Contact

Reach me at b343095@gmail.com or send a note below.