Projects
Built enterprise-grade speech-to-text pipelines and LLM applications using Whisper large-v3, PyTorch, FastAPI, and AWS EC2. Delivered real-time transcription APIs, domain-adapted chatbot interfaces, and end-to-end AI assistant systems for large-scale client deployments.
Impact: moved AI initiatives from exploration into production-focused delivery pipelines.
Engineered a multi-stage generative AI pipeline combining Whisper transcription, GPT-4 summarization, and LangChain agents to automate action-item extraction and follow-up drafting. Reduced post-meeting documentation time from hours to minutes across enterprise teams.
Impact: reduced manual documentation work and accelerated meeting turnaround.
Built a retrieval-augmented generation system using sentence-transformers, FAISS vector indexing, and cosine similarity ranking to match candidate profiles to job requirements. Improved screening precision for recruitment workflows handling thousands of applications.
Impact: increased candidate-job fit accuracy for hiring teams.
Developed visual search, item recommendation, and auto-tagging pipelines for fashion e-commerce using CLIP embeddings, ResNet feature extraction, and Pinecone vector DB. Improved catalog discoverability and reduced manual tagging overhead across large product inventories.
Impact: improved engagement and streamlined catalog operations.
ML-powered FPL prediction system using XGBoost regression, time-series feature engineering, and scikit-learn pipelines trained on historical and live match data to forecast player performance and automate team selection decisions.
Impact: enables data-driven FPL decisions with real-time predictive insights.
Contributed to the LangChain MCP adapter ecosystem for Claude integration, and published 15+ reproducible Kaggle notebooks covering PyTorch deep learning, NLP pipelines, and production ML engineering patterns used by thousands of practitioners.
Impact: extended open-source tooling and community learning resources.