Building intelligent systems at the intersection of Machine Learning, NLP, and Software Engineering.
I'm a 3rd-year M.Tech (Integrated) CSE student at SSN College of Engineering. I work across the full stack — from systems-level networking tools and IoT hardware to fine-tuned language models and deep learning pipelines.
My research has been recognised at MediaEval 2025 (3rd place, GI Imaging VQA) and DravidianLangTech 2026 at ACL (Top 10, Tamil NLP). I'm preparing for internships and further publications.
I'm drawn to hard problems — routing anomalies, pixel-level change detection, abusive language in low-resource languages. I care about getting the details right.
† School-level assessment only — AY 2020–21 Class X students were declared all-pass; no government board exam was conducted.
Applied BLIP-2 to the Kvasir-VQA-x1 dataset for VQA on endoscopy images, generating answers with explanations, attention maps, and confidence scores. Achieved 3rd place internationally.
Fine-tuned IndicBERT for Tamil abusive language detection with class weighting and threshold optimisation. Achieved Macro F1 of 0.8001, ranking in the top 10 of the shared task.
ElasticNet model with feature engineering, EWM imputation, and 10-fold cross-validation. Trading signals optimised by Sharpe ratio — final score ~3.99.
CLI tool comparing two PDF assignments across text (cosine, Jaccard, Dice n-grams), code (AST-level Python normalisation), visuals (perceptual hashing + tile matching), and document structure. Three-stage extraction pipeline; runs fully offline.
IoT crash-detection using Raspberry Pi, sensors, GPS, and GSM. Dispatches emergency alerts on collision with SOS button and environmental monitoring. IFSP collaboration.
Full-stack legal platform with a Mistral chatbot fine-tuned on Indian law data. Graph-native Neo4j backend for relationship-aware queries. Django API + Next.js frontend.
Scans QR codes, extracts and traces URLs, and classifies threats as safe/suspicious/malicious using heuristic rules and a trained ML classifier.
Platform for reporting and reclaiming lost items with auth, image-based posting, and a claim workflow. Vue.js frontend with Spring Boot REST API and MongoDB.
Detects routing loops via TTL variation and packet timing analysis. Real-time Flask dashboard with configurable alerts and anomaly visualisation.
Segmentation model on the Inria aerial dataset with CLAHE preprocessing. Generates pixel-level change detection maps from before/after imagery.
Automated seat allocation with student preference handling and matrix-based logic. Deployed on PythonAnywhere using SQLite; PostgreSQL locally.
Compared SGD, Momentum, Adam, RMSProp, Adagrad, Adadelta, LBFGS, and Photon on Rosenbrock, Rastrigin, Ackley, and Himmelblau benchmarks. Adaptive optimisers proved most stable.
Open to research collaborations, internship opportunities, and conversations about ML, systems, or software engineering.
Best reached by email — I try to respond within a day.