Production-Grade PDF RAG Chatbot
• Developed a modular, production-ready PDF RAG chatbot using FastAPI, LangChain, FAISS, and
Mistral-7B-Instruct, enabling semantic search and QA over uploaded documents
• Engineered an end-to-end pipeline with 95 percent chunk retrieval accuracy, covering PDF parsing, embedding
(MiniLM-L6-v2), vector indexing, and response generation via LLM.
• Integrated LangChain tools: RetrievalQA, ConversationalMemory, and custom agents for summarization + QA,
improving response quality and user context retention.
• Delivered both API and Streamlit interfaces with clean architecture and metadata-based filtering; reduced
average query latency by 30–40
Formula 1 Grand Prix Podium Prediction Web App
• Built a Flask-based web app to predict F1 race podium finishers trained on over 10,000+ qualifying session data
• Designed and optimized a TensorFlow-based neural network pipeline with a train-test split and Cross-Validation
to achieve a validation accuracy of 95.6 percent
• Conducted in-depth Exploratory Data Analysis (EDA), enhancing model performance (F1-score: 0.972)
• Deployed a responsive and interactive frontend with dynamic podium visualization for an engaging UI on render
Fresher Data Analysis
•Developed and executed comprehensive data analysis projects using Python, focusing on educational metrics like student performance, family
income impact, and CGPA distribution.
•Utilized Jupyter Notebooks, Excel, and advanced statistical methods to deliver actionable insights, documented in detailed reports and video
demonstration
Network_Anomaly_Data_Analysis_IITG
○ Developed a RandomForest model to classify 86,845 network activity records, identifying "Neptune" attacks versus normal activities. Trained
the model to predict outcomes for 21,712 test entries.
○ Achieved high accuracy of 99.2238 F1 score.
Heartbeat and Body-Temperature monitoring device
Developed a heartbeat and body temperature monitoring device present challenges such as ensuring accuracy under varying conditions, optimizing power consumption for long-term usability, and ensuring scalability and interoperability for comprehensive health monitoring.
With continued refinement, this device could empower individuals to manage their health more effectively while enhancing healthcare system efficiency
Algorithmic Trading System
Built a Python-based NIFTY 50 trading framework with RSI, SMA, and MACD indicators, OHLCV data ingestion, ML/DL predictive models, walk-forward validation, and a FastAPI plus Render deployment with live signals and portfolio analytics. Backtested with Sharpe 1.2, 55% win rate, and 8% drawdown.
