Featured Projects
LLM-Powered Documentation Assistant
Built a RAG-based Q&A system using LangChain, GPT-4, and ChromaDB to answer questions from large document sets.
Telemetry Prediction System
Designed a FastAPI-based system for real-time predictions with scheduled model training and PostgreSQL integration.
Cloud MLOps Pipeline
Automated ML model deployment pipelines on Azure Kubernetes Service using Azure DevOps and MLflow tracking.
LLM-Powered Documentation Assistant
A scalable, intelligent document Q&A system using LangChain, LLMs, and Vector Databases. Designed to extract accurate answers from technical documentation using Retrieval-Augmented Generation (RAG).
Problem
Needed a smart internal assistant to help employees query large volumes of technical product documentation and manuals. Traditional keyword-based search was ineffective, leading to wasted time and inaccurate answers.
Solution
- Developed a RAG pipeline using LangChain and OpenAI GPT.
- Indexed document corpus using ChromaDB (VectorDB) for fast similarity search.
- Deployed as an interactive web app using FastAPI + Streamlit.
- Chunked and embedded documents using sentence-transformers.
- Added logging and observability for LLM query effectiveness.
Architecture
The system uses a VectorDB (ChromaDB or FAISS) for document retrieval, passes results through LangChain’s RAG chain, and uses OpenAI’s GPT-4 to generate natural answers. The final output is rendered in a responsive UI.
Tech Stack
Impact
- 90% improvement in document search time.
- Enabled employees to self-serve technical Q&A with high accuracy.
- Set the foundation for multilingual support and feedback loop training.
Real-Time Telemetry & Prediction System
A real-time telemetry system built to predict asset safety using live metrics like setpoint, SOC, and temperature. This FastAPI-powered project handles model training, prediction, and asynchronous database operations, and is designed for future integration with Kafka and Redis for scalable streaming.
Key Features
- Async FastAPI endpoints for real-time data ingestion
- RandomForestClassifier for safety prediction
- Endpoints for submitting data, training model, and getting predictions
- Model serialization with
joblib - Support for dynamic threshold alerting and inference scheduling
Tech Stack
Scalability & Monitoring
Designed to support horizontal scaling with container orchestration (e.g., ECS, Kubernetes). Plans include Kafka integration for telemetry streaming, TimescaleDB for time-series data, and a background task scheduler for automated retraining and alerting.
Example Setup
# Setup Environment
conda create --name mlops_env --file requirements_challenge.txt
conda activate mlops_env
pip install -r pip-requirements.txt
# Run API locally
uvicorn predictive_api:app --reload
Use Case
Designed for use in energy, IoT, and industrial applications where real-time safety and performance monitoring are critical. Future enhancements include weekly/monthly reporting via unified endpoints and multivariate telemetry statistics.
View on GitHub
Cloud MLOps Pipeline Automation
Automated ML lifecycle using Azure DevOps, MLflow, and AKS for continuous deployment of scalable machine learning systems.
Project Overview
This project demonstrates a full MLOps pipeline automation leveraging Azure DevOps for CI/CD, MLflow for experiment tracking and model registry, and Azure Kubernetes Service (AKS) for scalable model deployment. The pipeline supports automated training, validation, and deployment workflows to streamline machine learning operations in production.
Architecture & Technologies
- Azure DevOps: Pipelines for continuous integration and delivery.
- MLflow: Experiment tracking, model versioning, and registry.
- Azure Kubernetes Service (AKS): Container orchestration and scalable deployment.
- Docker: Containerizing ML models and services.
- Python: Training scripts and API integration.
- Terraform: Infrastructure as code for resource provisioning.
Results & Demo
The pipeline successfully automates the model lifecycle from development to production with reduced manual intervention, ensuring robust deployment and monitoring.
Demo and screenshots will be added here soon.
View on GitHub