# Projects — Kuldeep Kumar Mishra

## 1. Agentic Hybrid RAG Chatbot (Flagship Project)
Type: AI Engineering / Multi-Agent System
Status: Deployed on Hugging Face

Description:
A sophisticated multi-agent chatbot system that combines Retrieval-Augmented Generation (RAG) with intelligent agent routing. The system uses multiple specialized agents coordinated through an intelligent orchestration layer to handle diverse user queries with high accuracy.

Tech Stack:
- Language: Python
- Framework: Flask, LangChain, LangGraph
- LLM: LLaMA 3.1 8B via Groq API
- Vector DB: ChromaDB
- Keyword Search: BM25 (Okapi BM25)
- Memory: SQLite (persistent conversation memory)
- Retrieval: Reciprocal Rank Fusion (RRF) — hybrid of vector + keyword search

Agent Architecture:
- RAG Agent: Answers questions from indexed documents using hybrid retrieval
- Web Search Agent: Searches the internet for real-time information
- Math Agent: Solves mathematical problems and equations
- Memory Agent: Retrieves and uses past conversation context
- General Agent: Handles general knowledge questions
- Router Agent: Intelligently routes each query to the correct agent

Key Features:
- Hybrid retrieval combining semantic vector search + BM25 keyword search via RRF
- Persistent conversation memory using SQLite
- PDF document indexing and retrieval
- Streaming responses for real-time interaction
- Multi-agent coordination via LangGraph

Deployment: Hugging Face Spaces (publicly accessible)
GitHub: Available on kuldeepmishra92

---

## 2. Skin Cancer Classification System
Type: Computer Vision / Medical AI
Status: Deployed on Hugging Face

Description:
A medical image classification system that uses deep learning to classify skin lesions into seven diagnostic categories. The system assists in early detection of skin cancer from dermoscopy images.

Tech Stack:
- Architecture: Google Vision Transformer Large (ViT-Large)
- Dataset: HAM10000 dermoscopy dataset (10,000+ images)
- Framework: PyTorch, Hugging Face Transformers
- Deployment: Hugging Face Spaces + Model Hub

7 Classification Categories:
1. Melanocytic Nevi (nv)
2. Melanoma (mel)
3. Benign Keratosis-like Lesions (bkl)
4. Basal Cell Carcinoma (bcc)
5. Actinic Keratoses (akiec)
6. Vascular Lesions (vasc)
7. Dermatofibroma (df)

Performance:
- Classification Accuracy: Over 92%
- Model: vit-large-skin-cancer-ham10000 (published on Hugging Face Model Hub)

Deployment: Hugging Face Spaces and Model Hub (publicly available)

---

## 3. Airline Passenger Satisfaction Prediction System
Type: Machine Learning / Tabular Data
Status: Deployed as Flask Web Application

Description:
A machine learning system that predicts whether airline passengers are satisfied or dissatisfied with their flight experience based on various features including service quality, flight details, and passenger demographics.

Tech Stack:
- Language: Python
- ML Framework: LightGBM
- Hyperparameter Tuning: Optuna
- Deployment: Flask web application
- Other Libraries: Scikit-learn, Pandas, NumPy

Approach:
- Evaluated multiple algorithms: Logistic Regression, Random Forest, XGBoost, LightGBM
- Selected LightGBM as best performing model
- Applied Optuna for automated hyperparameter optimization

Performance:
- Final Model Accuracy: Over 96%

Key Features:
- Interactive web interface for real-time prediction
- Feature importance visualization
- Model explainability

---

## 4. World University Rankings Platform
Type: Data Science / Analytics
Status: Published on Hugging Face

Description:
A data analytics and visualization platform for exploring and analyzing world university rankings data. Enables users to filter, compare, and understand ranking trends across institutions and countries.

Deployment: Hugging Face Spaces

---

## Published Hugging Face Projects
1. Skin Cancer Classifier (Application)
2. Multi-Agent RAG Chatbot (Application)
3. World University Rankings Platform (Application)
4. vit-large-skin-cancer-ham10000 (Model)

All projects available at: huggingface.co/Kuldeepmishra3