Spandan Vishal

Project: RAG–Based Equity Research Assistant

• Built an AI-powered equity research tool using RAG with ChromaDB and OpenAI LLMs to extract insights from financial news and filings.
• Implemented hybrid retrieval, caching, and async ingestion, improving query relevance and cutting manual research time.
• Deployed on AWS (ECS + Bedrock) with containerization, secrets management, and monitoring for scalability.
• Technologies: Python, OpenAI GPT-4, RAG, ChromaDB, Streamlit, AWS, Docker

Project: Hallucination Detection in LLMs

• Developed an end-to-end hallucination detection system leveraging metric learning, prompt engineering, and fine-tuning on recent LLMs such as BERT, Mistral, SOLAR, and MIXTRAL.
• Achieved state-of-the-art accuracy of over 93 % in identifying factual errors in LLM outputs for contrastive hallucination detection task.
• Technologies used: Python, LLMs, Deep Learning, Docker

Project: Credit Risk Modelling

• Developed an end-to-end solution for multi class classification task.
• Performed EDA, preprocessing tasks and built pipeline and also deployed it. Predicted credit risk of customers involved.
• Technologies: Python, Machine Learning, Flask

Project: Agentic Stock Broker (Genpact AI-CoE Academy)

• Built an end-to-end agentic AI pipeline (LangGraph) with four autonomous agents for fundamentals, quantitative analysis, news/sentiment, and competitor comparison; agents include memory and tool-use and call external APIs for enrichment.
• Aggregated multi-agent outputs with an LLM to produce personalized, explainable stock recommendations aligned to user investment plans; engineered for production with caching, retries, dotenv secrets, and unit-testable modules.
• Technologies: Python, LangGraph, LLMs (GPT-5 mini), yfinance, Finnhub, Alpha Vantage, OpenFIGI, NewsAPI/Twinword, requests, numpy/pandas, python-dotenv.