AI-Powered App Builder | Python · LangChain · FastAPI · Vector DBs
B.S. in Computer Science, University of Maryland, College Park | ML Specialization | Minor: Mathematics | GPA: 3.60
Built and deployed an AI-powered fact verification assistant using LangGraph orchestration over multiple trusted sources (Google, PubMed, Arxiv, Wikipedia, Tavily). The system selects tools based on the category of the claim, fetches full articles when available, and summarizes them with respect to the claim to keep evidence focused and concise. Users enter a factual claim, and Truth Chain delivers a verdict: True, False, or Unverifiable with reasoning. Evaluated on a curated dataset of 32 factual claims, achieving 87.5% accuracy with GPT-4o and 84.38% with GPT-4o-mini. Hosted on AWS EC2 with a clean Streamlit UI.
A local-first AI research assistant that indexes research PDFs into ChromaDB for semantic search with page-level citation metadata. Users upload papers and query via a Streamlit UI, retrieving top-K passages with sources/pages for accurate referencing. Evaluated on 12 queries with 91.7% Recall@5 and ~79% coverage. Built with Python, Streamlit, ChromaDB, Sentence-Transformers, PyMuPDF, and LangChain, showcasing a full pipeline from PDF → Chunk → Embed → Query → Retrieve → Display.
Built and deployed a dual-mode AI retrieval assistant using LangChain (precise matches) and LangGraph (intelligent fallback) over Ayurvedic remedy texts. Users enter symptoms, body type, and preferred remedy type, if no exact match exists, the system broadens its search adaptively, delivering remedies with 94% recall across 50 cases. Hosted on AWS EC2 with a user-friendly Streamlit UI.
Built an AI-powered email extraction and semantic search assistant using a hybrid C# + Python stack. Connects to mailboxes via IMAP, downloads .eml files by keyword/date, parses and embeds content using SentenceTransformers. Local SQLite + vector DB enables natural language search via a WinForms UI. Users can query emails like “monthly financial report” and get ranked results with cosine similarity. Evaluated on 60 emails across 5 query types, achieving ~67% Recall@10 with high top-K precision. Local-first, secure, and extensible for offline or enterprise use.
Email: gayatrichougule479@gmail.com
LinkedIn: linkedin.com/in/gayatri-chougule