01 — Projects
Projects
Things I've built: agents, retrieval systems, interpretability tools and data apps. Each one links to its code, and to a live demo where there is one.
-
2026 TickerTape Agents that read 10-K filings, graded against the SEC’s own XBRL data: 99.8% figure accuracy on 54 real filings, with restatement detection and MD&A claim checking. Code for TickerTape on GitHub (opens in a new tab) Live demo of TickerTape (opens in a new tab)
-
2026 Citeline A retrieval-augmented research assistant over 131 ML papers that answers only from retrieved excerpts, with citations. 77% top-1 and 90% top-k retrieval accuracy. Code for Citeline on GitHub (opens in a new tab) Live demo of Citeline (opens in a new tab)
-
2026 AttentionForge Studio A full-stack Transformer interpretability studio: raw PyTorch attention diagnostics, explained by Claude. Code for AttentionForge Studio on GitHub (opens in a new tab) Live demo of AttentionForge Studio (opens in a new tab)
-
2026 DriftLens Detects ML model failures with drift analysis and feature importance, then explains the root cause using RAG. Code for DriftLens on GitHub (opens in a new tab)
-
2025 QuantSight A live stock-trend prediction app in Python, with its own market-data pipeline. Code for QuantSight on GitHub (opens in a new tab)
-
2025 PM Speech Sentiment Analyzer An interactive Streamlit app that reads Prime Minister Shehbaz Sharif’s speeches with NLP: sentiment, emotions, trends over time and emotion radar charts. Code for PM Speech Sentiment Analyzer on GitHub (opens in a new tab)