- Rich Text Doc RAG for arXiv Papers -

Python, FastAPI, DynamoDB, CLIP, OpenAI Embedding Model

Implementing a specialized version of Retrieval-Augmented Generation (RAG) dealing with academic arXiv papers containing complex structures like images, tables, texts and formulas to efficiently retrieve and generate content based on a specific topic and time range making it a valuable tool for students/researchers seeking appropriate content from the vast repository of arXiv papers.

- Kubernetes-based Data Processing Pipeline -

Kubernetes, Docker, neo4j (Graph DB), Python

Engineered a highly scalable data processing pipeline using Kubernetes pods for Apache Kafka and Neo4j that processes 100,000+ NYC taxi records in near-real-time, reducing spatial analytics query time by 60% and enabling real-time traffic pattern identification.

- NetCDF File Merger -

Rust, Rocket

A Rocket-based HTTP server that merges global attributes and variables from two NetCDF4 files efficiently, entirely in memory. No data ever touches disk.

- TalkDocAI -

Python, FastAPI, OpenAI gpt-4.1

A simple system that lets users upload a document, answer questions through a chat interface, and download a completed version of the document.

- Automated Sales Lead Enrichment Tool -

Python, FastAPI, OpenAI gpt-4

A full-stack pipeline that automatically enriches sales leads with real-world data and generates personalized outreach using AI. The system pulls in parallel from public APIs (Wikipedia, Google News RSS, BLS unemployment data, Open-Meteo weather), scores each lead 0–100 with GPT-4, and writes a tailored cold email — all written back to the sheet automatically. Built with FastAPI and a daily APScheduler job, with real-time Server-Sent Events (SSE) for live UI updates.

- AWS-based Image Processing Pipeline -

AWS Services (EC2, SQS, Lambda, S3, Autoscaling), PyTorch, FFMPeg, Docker

Part 1 - A cloud-native image processing application built on AWS that enables
asynchronous image processing at scale. The system leverages a
microservices architecture with seamless auto-scaling capabilities.

Part 2 - A serverless video processing and analysis system built on AWS Lambda
that automates video frame extraction and analysis. This pipeline
demonstrates modern event-driven architecture for efficient video
processing.

- TCR-Epitope Binding Affinity Prediction -

Computational Biology

Python, Transformer architecture, BERT model training

Understanding TCR-epitope binding is vital for advancing immunotherapy,
vaccine design, and personalized medicine, as it helps identify
targetable antigens and optimize therapeutic strategies for various
diseases.

The main goal of this project is to predict the binding affinity between
TCRs and Epitopes in order to understand whether they are likely to bind or any alternate measures are to be taken.