I ❤️ building cool stuff 👨‍💻 💻 Link to heading

Here are some of my projects & publications Link to heading

• Optimize Visual Shopping Journey with Embedding-based Retrieval in Pinterest Closeup | SIGIR'25

  Optimizing embedding-based retrieval for personalized product Pins on Pinterest. Introduces Shopping Priority Corpora built from probability models over a dynamic billion-scale inventory, paired with multimodal multi-task contrastive learning to balance relevance and engagement. Deployed in production with significant metric gains.

• Revisiting Open Domain Query Facet Extraction and Generation | ICTIR'22

  Open domain query facet extraction via sequence labeling and extreme multi-label classification, outperforming prior baselines. Released the Faspect toolkit.

• Probe — Personal Research MCP Server

  A local MCP server that ingests research sources (papers, filings, Substacks, tweets, notes), extracts structured information, and connects new content to existing knowledge via RAG. Exposed to Claude Code so it can query the index and evaluate research theses on demand. Built end-to-end with Claude Code — no code written, read, or reviewed by me; my contribution was harness engineering, prompting, and agent orchestration.

• Maruna Bot: An extensible retrieval-focused framework for task-oriented dialogues

  A retrieval-focused Task-Oriented Dialogue System for cooking and DIY tasks with speech and multimodal interfaces. Built for the Alexa Prize Taskbot Challenge 2021.

• Drink bleach or do what now? Covid-HeRA | ICWSM'22

  A dataset for assessing the severity of COVID-19 misinformation on social media and detecting high-risk fake news and refuted claims.

• Senior Thesis: Deep Patient for Summarizing Health Records

  EHR embeddings via stacked denoising autoencoders, outperforming raw-data and other dimensionality-reduction baselines.