Associate AI Engineer — R&D @ MathCo (Jun 2025 – Present)

• Deployed end-to-end QA automation using Playwright-MCP (REST API + cronjobs + monitoring dashboard)
• Engineered LangGraph agent orchestration — stateful reasoning, dynamic tool selection, memory-aware execution
• Contributed to Text-to-SQL pipeline and automated PowerPoint generation system for client delivery

AI Intern — R&D @ MathCo (Mar 2025 – Jun 2025)

• Built production multi-modal RAG pipeline for NucliOS — PPTX, PDF, DOCX, TXT — with semantic chunking & hybrid retrieval (FAISS, ChromaDB)
• Pipeline now live across multiple client deployments and product features org-wide

Deep Learning Research Intern @ BrainLamp Solutions (May 2024 – Aug 2024)

• Researched source-target distribution alignment for zero-shot & few-shot transfer learning
• Studied adversarial training, optimal transport & contrastive objectives for cross-domain generalisation
• Work contributed directly to a Springer publication (Artificial Intelligence Review, 2024)

Zero-Shot Plant Disease Classification with Semantic Attributes Artificial Intelligence Review, Vol. 57, Article 305 — Springer Nature (Open Access) · Sep 2024

Concatenated CNN / ViT feature vectors with semantic attribute vectors to classify unseen plant diseases. Achieved 99.81% accuracy with ResNet-18 on cross-species transfer, outperforming InceptionV3 + TL (88.62%) by +11 pp.

Entropy-Guided Transfer Learning for PM2.5 Estimation from Satellite AOD IEEE · 2025

Applied Shannon entropy weighting to suppress negative transfer from Delhi → Kolkata & Guwahati. Benchmarked RF, XGBoost, SVR, MLR baselines. Lifted PM2.5 prediction R² from 0.88 → 0.98.

Spatial Statistics Conference 2025 — Poster Accepted

Designed a hybrid Active RL framework using MODIS satellite AOD data and Gaussian Process Regression, with a Q-learning agent selecting optimal sensor locations. Outperforms random and greedy sampling baselines.

Reinforcement Learning Gaussian Processes Kriging MODIS

Research Project · 2024

Generalised air quality forecasting across Indian cities using LSTM + Transfer Learning. Addresses the gap where 80% of Indian cities have only one CPCB monitoring station.

LSTM Transfer Learning ARIMA Time Series Climate AI

Smart India Hackathon 2023 — Top 6 / 200 Teams

End-to-end multilingual NLP chatbot for citizen grievances across 5 Indian languages using BERT + PyTorch. Automated complaint routing and real-time status tracking.

BERT PyTorch NLP Multilingual Intent Classification

Springer · Artificial Intelligence Review · 2024

Zero-shot framework combining CNN/ViT features with semantic attributes — 99.81% accuracy on cross-species transfer. Published in Springer Nature.

Zero-Shot Learning Vision Transformers CNNs ResNet Semantic Attributes