steramlit

This project employs machine learning and synthetic dataset to predict the peak equivalent stress imposed on a CFRP wrapped HDD overbend

🧾 OCR AI Streamlit App – Text Extraction from Images using Groq Vision LLM

A robust deep learning pipeline leveraging Convolutional Neural Networks (CNNs) to accurately detect and classify brain tumors from MRI scans with high performance and scalability.

An AI-powered research assistant that automates deep-dive information gathering. Orchestrates Planner, Worker, and Critic agents to produce data-driven, fully cited executive briefs.

Educational Streamlit apps powered by Large Language Models to support teaching, learning, and formative assessment.

Build a Retrieval-Augmented Generation (RAG) chatbot for local PDFs/texts using ChromaDB, LangChain, and Ollama with a Mistral model—completely offline, no OpenAI API needed.

LSTM Stock Prediction App for 100 Day prediction

AI-integrated application that predicts house selling prices based on selected variables.

Developed a computer vision-based crowd detection system for real-time people counting and density estimation using deep learning.

Implementation of a Wasserstein Generative Adversarial Neural Network with PyTorch

End-to-end hybrid RAG system tackling the "Lost in the Middle" phenomenon in academic PDFs using layout-aware parsing, BM25 + dense vector fusion (RRF), FlashRank re-ranking, and async Map-Reduce synthesis.

Production-ready Hybrid RAG Assistant for the Scrum Guide (2020) using FAISS, BM25, Cross-Encoder reranking, and Groq (Llama 3.3). Decoupled ingestion (Colab) from inference (Streamlit Cloud)

Predict Sri Lankan car prices from listing attributes (make, model, year, mileage, transmission, fuel type, engine CC, and simple condition flags extracted from ad text).

Multimodal E-Commerce Intelligence Platform combining GenAI Agents, Computer Vision (PyTorch), PySpark ETL, and Causal Inference.