Introduction

Hello,
My name is Hemanto Bairagi , I'm a Professional Software Engineer.

Bridgewater Associates – Take-Home Assessment (2025) - AI Workflow System Design Project

Designed and implemented a fully functional workflow builder and executor using ReactFlow and LangGraph.js. Built custom CRUD logic for workflows and nodes; persisted execution history to serverless PostgreSQL (Neon). Integrated OpenAI's Chat API for live LLM responses; managed dynamic graph state with modular JSON schemas. Delivered responsive frontend and DRY backend logic with real-time feedback, validation, and debugging.

Universal RAG Co-Pilot Generator (Bravura AI)

Developed a domain-agnostic RAG framework that transforms unstructured sources (PDFs, HTML, text) into structured SQL and vector databases. Enabled rapid chatbot and co-pilot creation for automation, finance, legal, and compliance domains. Original use case involved training automation engineers; later expanded into alpha signal generation with domain-specific LLMs.

Mathematical R&D Optimization Framework (Bravura AI)

Created a method to accelerate GenAI product development using core mathematical and heuristic optimizations. Reduced experimentation cycles and improved model performance alignment with KPIs. Focused on practical PyTorch-level tuning, hyperparameter convergence, and evaluation loop efficiency.

Adamas Audio - TTS Engine & Audiobook Platform

Built a comprehensive machine learning application designed to create customizable audiobooks using voice and optical character recognition. Developed a Transformer-based speech synthesis system supporting English, French, and Hindi. Used machine learning to perform voice synthesis and OCR to create synthetic voice engines that turn PDFs into customizable audiobooks. Containerized via Docker and deployed on IBM Cloud for scalable audiobook generation.

Fraud Detection ML Pipeline & Auto-Dashboard System (Bravura AI)

Designed an end-to-end ML platform for fraud detection using credit card datasets as a base use case. Built a modular AI adapter layer using LLMs (Claude API, LangChain) to intelligently match and transform arbitrary tabular schemas. Integrated Power BI to auto-generate dashboards using natural language queries and schema-aware data access.

Sentiment Analysis API & UI (Independent Project)

Developed a full-stack sentiment classification platform using Python, FastAPI, and Hugging Face Transformers. Fine-tuned a DistilBERT model on labeled tweet sentiment datasets for binary classification. Built a clean React frontend for users to input text and view live sentiment analysis with probability scores.

Deep Learning SAR Image Classification System

Built production-ready SAR (Synthetic Aperture Radar) ship detection system with full-stack web interface and REST API. Implemented custom CNN architectures using TensorFlow/Keras for maritime vessel detection and classification in satellite imagery. Created geospatial processing capabilities using GDAL, rasterio, and geopandas.

Kubernetes-Deployed Geometric Calculator Microservice

Built a scalable microservice for geometric calculations (area, perimeter, volume) deployed on Kubernetes. Implemented FastAPI backend with Pydantic data validation and automated OpenAPI documentation generation. Created containerized architecture with multi-stage Docker builds.

Enterprise Q&A System with RAG Architecture

Developed a production-ready Retrieval-Augmented Generation (RAG) system for intelligent document-based question answering. Built React 18.2.0 frontend with TypeScript, Tailwind CSS, and real-time WebSocket communication for streaming AI responses. Integrated Claude 3.5 Sonnet API for high-quality response generation.

MLOps BERT Sentiment Analysis Pipeline

Built end-to-end MLOps pipeline for BERT-based sentiment analysis with automated training and deployment. Implemented continuous integration for model training, validation, and versioning using MLflow. Created Apache Airflow DAGs for automated data processing, model retraining, and performance monitoring.