All Projects

Explore all my projects and their details

Tax Concierge AI - Intelligent Routing System

Architected an agentic RAG pipeline with query routing, classifying complexity via semantic and LLM-based scoring. Resolved simple queries through hybrid retrieval with BM25 and pgvector, and escalated complex queries to domain experts using Llama 3.3.

AI Model Quality & Performance Challenge

Built a performance UI that turns internal .xlsx perf projection sweeps (Models A to K across 7 traffic profiles) into actionable views: a go/no-go signal for customers and PMs, and a sanity-check view for engineers, with live upload and side-by-side model comparison. Built a benchmark compression pruner inside evalscope as an upstream-quality extension, reducing coding (LiveCodeBench), long-context (AA-LCR), and multimodal (MMMU) benchmarks to the smallest sample set that still gives a useful good-or-not signal.

School Recommendation System

Built a recommendation system that matches students to suitable schools based on their preferences and profile data. Implemented data-driven scoring to rank and surface the most relevant school options for each user.

Automotive Sales Performance Dashboard

Built a Tableau dashboard with 12 interactive worksheets and advanced filters, enabling 50% faster identification of key sales performance patterns. Automated data wrangling in Tableau Prep Builder, achieving 100% elimination of missing and invalid data across all reporting fields.

Revenue Analytics Dashboard

Migrated 10000+ Adidas sales records to MySQL, automating Tableau real-time refresh and eliminating all manual data update workflows. Developed a revenue Tableau dashboard identifying sales patterns projected to drive 25% improvement in overall revenue performance outcomes.

Homomorphic Encryption for Secure Evaluation of Images

Developed a scalable Fully Homomorphic Encryption (FHE) pipeline that allows for model inference on encrypted data without the need for decryption.

Auto Documentation - Code to Docs with Live Preview

Built a tool that turns source code into clean, structured documentation with a live preview, helping developers generate readable docs directly from their codebase. Designed an intuitive editor and preview workflow so changes to code reflect instantly in the rendered documentation.

AI-powered CRM Dashboard

Developed a CRM dashboard for 300+ student records, centralizing communication logs, progress tracking, and engagement history. Generated AI-driven alerts for at-risk students, helping advisors prioritize outreach through data-driven insights.

AI Assistant for Project Management

Architected an autonomous Enterprise AI Assistant using LangChain and Gemini 1.5 Flash, integrating Text-to-SQL (PostgreSQL) and RAG (pgvector) to provide real-time insights across 15k+ employee records and project datasets. Engineered a production-grade LLM pipeline featuring LangSmith for observability and LLM-as-a-Judge evaluation frameworks, reducing manual audit overhead while ensuring 100% adherence to security protocols via LLM Guard integration.

Visibility Prediction System for airports using DeepAR model

Conceptualized and built an end-to-end ML workflow on AWS to forecast airport visibility hourly using SageMaker DeepAR models (NWP). Attained 92% accuracy in predicting visibility using DeepAR (Auto Regressive model) after evaluating and comparing various models including LSTM, ARIMA, and others

Oil Spill Detection using MCAN model

Designed and implemented an AI solution using Multiscale Conditional Adversarial Network (MCAN) and subpixel optimization for oil spill detection from Synthetic aperture radar (SAR) imagery aiding environmental protection and emergency response.

Flood forecasting using Deep learning, IoT, and GIS

Pioneered an AI system using a hydrological model (LSTM) on GloFAS data to predict and visualize high flood-risk areas. Integrated real-time IoT sensor data to forecast flood levels and built interactive maps and apps to alert in case of flood hazards.

Predicting High-Impact Weather Events ocean disasters

Formulated a generalized machine learning model using DGMR to predict probabilities of high-impact weather events and ocean disasters within a 5 to 72-minute timeframe (nowcasting).

Wireframes For an Asset Tracking System

Created a modular, role-based design enhanced by validation logic, authentication, notifications, dashboard views, and cloud deployment to create a user-friendly yet secure system.

Nutpam 2021 - Tech Fest

Orchestrated end-to-end web development and IT efforts as Chief Web Developer for Nutpam with a team of developers to swiftly build and deploy the visually appealing, responsive website for the tech conference and achieved a page-load speed of 53 milliseconds.

Nutpam 2022 - Tech Fest

Led the technology efforts as Web Lead for the Nutpam 2022 tech fest, the annual symposium at Sairam Institute of Technology.Developed the website project end-to-end, including planning, designing, developing, and launching the responsive website with a performance score of 94% from Google PageSpeed Insights