Int. M.Sc. Data Science student at Amrita Vishwa Vidyapeetham (2022–2027).
Building end-to-end ML pipelines across Sports Analytics, Data Analytics,
Explainable AI, and Large Language Models.
My passion for cricket and football drives my research — applying Graph Attention Networks,
Transformers, RAG pipelines, and SHAP explainability to extract real, actionable insights from raw match data.
GAT + BiLSTM + cross-attention Transformer for IPL T20 outcome prediction. 225K deliveries, 950 matches. Ball-by-ball win probability & Player Impact Score with SHAP.
Multi-task RoBERTa with novel Sentiment Incongruence Auto-Labeler. F1-Macro 0.977, AUC 0.997. Cross-domain gains on 3 benchmarks. LIME token-level rationale extraction.
Classifies Attacker/Midfielder/Defender from wearable sensor data. TCN+Transformer achieves 99.24% accuracy. LOSO 98.89%±0.42%. SHAP: ankle accel → Attacker.
UCL match summaries via RAG + LLaMA 3.1 + SHAP. Self-curated dataset: 189 matches, 142 columns. Sentence-BERT + FAISS: cosine similarity 0.903 vs 0.373 baseline.
Multi-league framework across 5 European leagues. ELO + rolling form + betting probabilities. 33 features, 25,979 records, 259 teams. Gradient Boosting: +0.073 F1 vs baseline.
End-to-end manufacturing audit dashboard on 3,240 production batches. 10 KPIs incl. OEE (85.37%), X̄-R control charts (±3σ). Batch Risk Scorer with SHAP, ROC-AUC 0.7417.
Agentic study assistant with Observe→Think→Act loop using Groq + LLaMA 3.3-70B. Finds curated resources and generates personalized exam study plans with live progress tracking.
AI goalkeeper trained via PPO in a 2D Pygame simulation. Rule-based threat assessment + shot prediction. State machine: IDLE→TRACKING→READY→DIVING→RECOVERING. Improves each game.
AI platform for students with ADHD & dyslexia. T5 PDF summarization, spaCy quiz generation, TTS, multilingual translation in 7 Indian languages. Django + PostgreSQL backend.
3-class fatigue prediction from 2.87M wearable sensor readings. Karvonen HR labeling, 88 features per window. Personalized RF: 97.87% accuracy. LOSO 97.96%±2.57%.