AI/ML Projects

Production-grade multi-agent system with retrieval-augmented reasoning and a modern chat interface

Built a sales-focused agentic architecture orchestrated via LangChain with RAG over a PostgreSQL + pgvector knowledge base, FastAPI services, and a Next.js/TypeScript chat frontend. Deployed using Docker on Render with environment-based configuration and observability.

Key Features:

• LangChain multi-agent orchestration with tool-use and guardrails
• RAG retrieval over domain knowledge using pgvector embeddings
• Next.js chat UI with streaming responses and contextual memory
• FastAPI backend services with typed models (Pydantic)
• Dockerized deployment on Render with secrets management

Tech Stack:

Production-grade NLP microservice for real-time emotion classification with full-stack deployment

Built a complete AI application featuring a fine-tuned DistilRoBERTa Transformer for emotion classification, deployed as containerized microservices on Google Cloud Platform with automated CI/CD pipelines.

Key Features:

• Fine-tuned DistilRoBERTa Transformer for emotion classification
• FastAPI backend with async inference and performance caching
• React + Vite frontend with Tailwind CSS for real-time visualization
• Docker containerization with Cloud Run deployment
• Automated CI/CD via Firebase CLI and Google Cloud Build
• Environment-based CORS control and Pydantic validation

Technical Highlights:

• Low-latency, cloud-native AI product showcasing full-stack ML and MLOps
• Production-ready microservice architecture with scalable inference
• Integrated frontend-backend communication with real-time emotion outputs
• Complete DevOps pipeline from development to deployment

Tech Stack:

Sequence-based classifier achieving 91% accuracy and 0.96 ROC-AUC using engineered features from protein sequences and ML models.

CNN-based tumor classification in MRI with Grad-CAM interpretability and medical imaging preprocessing.

Natural Language Processing classifier to detect fake news using TF-IDF vectorization and Logistic Regression, with data leakage prevention techniques for improved model generalization.

Key Technologies:

Utilized clustering and dimensionality reduction techniques (PCA, K-Means, DBSCAN) to predict student performance and identify struggling students early for timely intervention.

Key Technologies:

End-to-end ML pipeline for diabetes risk prediction with class balancing techniques and hyperparameter tuning, transforming biomarkers into risk scores for early disease diagnosis.

Key Technologies: