Devroop Saha

navilaw-ai. AI powered legal research engine. The system is based on multi AI agentic RAG systems leveraging the power of Llama3 LLM

TexMCP. A small FastMCP-based Microservice that renders LaTeX to PDF. The server exposes MCP tools to render raw LaTeX or templates and produces artifacts

fastapi-scaffold. FastAPI Scaffold is a CLI tool to quickly generate FastAPI project structures with optional features like authentication, database integration, machine learning model setup, and Docker support.

YogaFix. YogaFix is a real-time Yoga Pose detection and feedback system built using Python, OpenCV, and Mediapipe for pose estimation. The system is served via a FastAPI backend that captures webcam frames server-side, processes them to detect poses, and provides real-time feedback over WebSocket connections.

LangChat. A RAG based QnA Application over the LangChain documentation.

CourseLens. A RAG Based smart search system, recommending courses on the basis of natural language based queries and keywords.

julius_ai. Julius AI is an advanced AI-powered platform that conducts complete technical interviews through real-time voice interaction, structured conversational assessment, and separate coding evaluation

Chirpy. It is an AI-powered integration for Sparrow, revolutionizing API management with context-aware chatbots and automated workflows. It leverages the power of RAG, Prompt engineering and Llama3 to streamline developer tasks and enhance API lifestyle.

HateSpeechDetect-text. In this project, I focused on benchmarking various machine learning models, deep learning architectures, and fine-tuned BERT-based models to evaluate their performance across multiple metrics

toxicity-prediction-ai-bionnovate-Visonx.py. A model to predict the toxicity of a given compound through its SMILES notation

FitVid. This is AI powered video based gym trainer, that monitors the form of gym exercises.

Bank-churn-prediction. Bank Churn Predicition using every type of classification Algorithm

House_Price_Prediction_Linear_Regression. This is a machine learning project based on predicting house prices by implementing Linear Regression

Boston-Housing-price-App. This is the MLCOE task 4 repository

agatha. Agatha is a NL2SQL-driven expense tracker API that leverages Llama 3.3 provided by Groq for efficient and fast data extraction and NL2SQL workflow. It allows users to either add transaction data to a SQL database or ask questions about their expenses, all via natural language.

Diabetes-Prediction-using-Logistic-Regression-beginner-level-. This project is based on the diabetes prediction model using Logistic Regression.

NovaAI. This multi-AI agent system has been developed to automate research and writing blogs. On the basis of the given topic as input, it researches and fetches all the content related to that particular topic and accordingly , it jots down a comprehensive blog or a structured document based on the fetched content.

Dinesh-FitChat. This is a RAG based chatbot that acts like a fitness trainer and gives fitness advice on training, diet or motivation.

Elevate. This project implements and visualizes various disk scheduling algorithms using the analogy of elevators in a building. The primary algorithms include FCFS, SSTF, SCAN, LOOK, and their circular versions.

Predicting-time-series-using-xgboost. This repository demonstrates the time series forecasting power of xgboost

MLCOE-TASK-2. MLCOE task-2

multiclass-classification-using-DecisionTrees. This is notebook is the implementation of Decision Trees for multiclass classification on the bean dataset.

Langchain-learning. This is fun little project where I used Mistral7b LLM via HuggingFaceHub and Langchain wrapper

leetode-contest-rating-prediction. I am developing a model which will predict the rating of a contestant based on his/her performance on the contest

voicebook. This project is a realtime voice booking agent that captures speech in the browser, streams it to a Node.js WebSocket backend, transcribes with Deepgram STT, invokes a Groq LLM, and returns low-latency audio responses synthesized by AWS Polly . it then automatically validates the user provided name, email and dates and then stores it on local CSV.

legal-reformatting. Python

whisper-hindi. Jupyter Notebook

agno. A lightweight library for building Multimodal Agents. Give LLMs superpowers like memory, knowledge, tools and reasoning.