R2AI is a semi-autonomous refactoring agent built using LangGraph. It analyzes and improves Python code structure using Ruff, refactors it via LLM, and supports human-in-the-loop review with feedback that can trigger iterative improvements.
Technologies Used
Python, LangGraph, LangChain, Ollama, Mistral, OpenAI API, Ruff, Poetry
Agent Cortex v2 expands the local AI assistant with new tools including: persistent long-term memory, Python code execution, and short-term conversational context. Built on LangChain and Mistral 7B, it enhances the original RAG, math, and web search tools
Technologies Used
Python, LangChain, Mistral, Ollama, ChromaDB, FAISS, DuckDuckGo Search, Python Code Execution, Poetry
Agent Cortex is a local AI assistant powered by LangChain and Mistral 7B. It combines: Document retrieval (RAG), Math solving, and DuckDuckGo web search. Runs fully offline (except for websearch)—no paid LLM APIs
Technologies Used
Python, LangChain, Mistral, Ollama, ChromaDB, FAISS, DuckDuckGo Search, Python Code Execution, Poetry
production-ready RAG (Retrieval-Augmented Generation) backend built with FastAPI, SentenceTransformers, and FLAN-T5. It demonstrates a modern approach to question-answering using both semantic search and language generation.
Technologies Used
Python, FastAPI, SentenceTransformers, transformers, torch, numpy, FAISS, FLAN-T5, Docker, Poetry
A lightweight computer vision project developed to deepen my understanding of image preprocessing, transfer learning, and end-to-end ML deployment with minimal cloud cost. It's built using transfer learning with CNN EfficientNetB2
Technologies Used
Python, TensorFlow, Keras, scikit-learn, EfficientNetB2, Transfer Learning, matplotlib, seaborn, firebase, google-cloud-aiplatform, Docker, FastAPI, Poetry
a portfolio project built to experiment with and deploy lightweight abstracted summarization models based on the T5 (Text-to-Text Transfer Transformer series) models. Designed to summarize multi-turn conversations (e.g., meetings, support chats)
Technologies Used
Python, PyTorch, T5, Hugging Face Transformers, Numpy, FastAPI, Firebase, Google Cloud Run, Google Cloud Registry, Docker, Poetry
A production-grade machine learning system that predicts whether a user is currently employed based on structured profile attributes. It uses a supervised binary classification model built with TensorFlow/Keras (Sequential API) — a deep feedforward neural network trained on structured tabular data.
Technologies Used
Python, TensorFlow, Keras, Pandas, Scikit-learn, Matplotlib, Seaborn, FastAPI, PostgreSQL, Cloud SQL, Google Cloud Run, Google Cloud Registry, Docker, Poetry