Software Engineer &
AI/ML Systems Builder.
Full Stack · Machine Learning · Distributed Computing · Databases
Featured
PicksAIStora | Full-Stack Analytics Platform
Architected an in-memory columnar query engine in Python processing 100K+ row datasets with sub-second filtering, grouping, and aggregation outperforming naive pandas by 3x on benchmarks. Exposed functionality via Flask REST APIs with JWT auth and RBAC, supporting secure multi-tenant querying with per-user schema isolation backed by PostgreSQL. Containerized all microservices with Docker Compose and automated CI/CD via GitHub Actions, maintaining 85% test coverage. Built a React dashboard with real-time chart rendering using Recharts, enabling non-technical users to query and visualize large datasets without SQL knowledge.
Human-AI Collaboration for Backend Text Generation: Dynamic Content Recommendation for Websites based on Keywords
Co-developed a keyword-driven content recommendation engine leveraging GPT-based AI pipelines to dynamically generate and personalize backend website copy, reducing manual authoring time by 60%. Evaluated system output quality against human-authored baselines through user studies with 40+ participants measuring coherence, relevance, and preference scores.
Experience
tap a skill pill to exploreFreelance Full-Stack Developer
Aadithya Cars
Reduced page load time 75% (3.2s → 800ms) via lazy loading, code splitting, and Brotli compression, raising Lighthouse Performance score to 95+. Replaced monolithic CSV ingestion with a chunk-based streaming pipeline, eliminating OOM crashes and achieving 4× throughput on 500K+ row inventory datasets. Integrated third-party valuation and financing APIs with exponential-backoff retry and circuit-breaker patterns, maintaining 99.5% uptime and cutting time-to-interactive by 40%. Improved SQL query performance 35% via composite index design and query plan analysis on PostgreSQL tables with 1M+ records.
Software Engineering Intern
Virtusa
Led a team of 4 interns to deliver an EMart Inventory Management System (Spring Boot, Angular, MySQL) — real-time warehouse/shelf stock tracking, expiry alerting 2–3 days in advance, and maker-checker approval workflow, reducing manual stock discrepancy errors by ~40%. Shipped 3 backend modules (Public Employment Center, Tamil Nadu EB Bill System, barcode generation) with a CSV ingestion pipeline validating SHA-256 hashes and auto-generating signed PDF invoices emailed to vendors, cutting invoice processing time from ~2 days to same-day.
Software Engineering Intern
Virtusa
Led a team of 3 interns to build and demo a full-stack e-commerce grocery application (Spring MVC, Angular 9+, MySQL) end-to-end in 6 weeks — covering product catalog, cart, checkout, and order management across 5 core user flows. Designed 8 RESTful API endpoints and a normalized MySQL schema with Spring Hibernate ORM; integrated Spring Security for auth, achieving <3s page load under simulated 50-user load during final demo review.
Research Assistant
R.M.K Engineering College
Contributed to the development of ML-driven applications under faculty supervision, supporting data ingestion, preprocessing pipelines, and backend integration. Implemented backend systems in Python integrating machine learning models for sentiment analysis, text generation, and medical image classification. Assisted in experimental evaluation, result analysis, and manuscript preparation, contributing to 3 accepted IEEE conference publications (7 total citations).
Software Development Intern
SAWBON
Built concurrent Go REST APIs using goroutines, channels, and connection pooling, sustaining sub-250ms p95 response times under 500 concurrent users with zero downtime. Implemented Redis cache-aside with TTL-based invalidation, reducing DB read load by 30% and cutting median API latency by 15% . Achieved 78% code coverage on critical service paths via unit and integration tests using Go's testing package.
Projects
Deep Neural Net IPL Predictor | Multi-Modal Win Forecasting
NewBuilt a multi-modal deep learning framework for IPL match outcome prediction achieving 80% test accuracy on 2025 IPL matches, integrating performance data from 13 cricket competitions across 235 players using league-quality weighting and recency decay. Designed a Dynamic OVR rating system (55-97 scale) with phase-specific scores for batting positions (Top Order, Middle, Finisher) and bowling phases (Powerplay, Middle, Death), feeding a 4-source weighted attention transformer with learnable fusion weights. Enhanced predictions with a 2-layer Graph Attention Network modeling batter-bowler confrontation asymmetries, combined with a 10,000-iteration Monte Carlo simulation engine yielding a Brier score of 0.24. Deployed as a production Flask API with SHAP explainability and sub-2-second latency.
GraphQL Task Management API
Designed a schema-first GraphQL API using FastAPI and Strawberry, eliminating REST over-fetching and reducing average payload size by 60% on deeply nested task/subtask queries. Integrated Redis DataLoader batching and OAuth 2.0 PKCE flow supporting 1000+ concurrent WebSocket connections at sub-100ms p99 latency under load testing.
Research & Publications
Automated Pneumonia Detection using DenseNet
NewCo-trained DenseNet-121 on the CheXpert chest X-ray dataset with augmentation and class-weighting to handle label imbalance, achieving 92% precision-recall AUC on the held-out test set. Benchmarked against ResNet-50 and VGG-16; DenseNet outperformed by 6% AUC via dense skip connections enabling richer feature reuse across layers.
Sentiment-Based Drug Recommendation System
Co-developed an NLP pipeline using fine-tuned BERT on domain-specific patient review corpora to classify sentiment and generate ranked drug recommendations at 88% classification accuracy. Evaluated against TF-IDF and LSTM baselines, demonstrating a 12-point F1 improvement from transfer learning with biomedical pretraining.
Stack
Click any pill to see every project, role and paper where I used it
Others
Contributor — PrintStruct
Open SourceContributed to PrintStruct, a Python CLI tool on PyPI that visualizes project directory structures while respecting .gitignore rules. Refactored project structure for clarity, eliminated code duplication to reduce technical debt, and improved README documentation to lower barrier for new contributors.
Microsoft Imagine Cup
HackathonBuilt a collaborative platform connecting researchers, doctors, and patients to share medical reports and treatment knowledge for rare disease research.
Shell.ai Hackathon: EV Charging Challenge
HackathonOptimized an EV charging network topology to remain robust under demographic shifts and meet dynamic customer demand.
Let's
Connect.
Whether you're looking to build scalable infrastructure, real-time data pipelines, or just want to chat about backend systems, my inbox is always open.