Pradyot Bathuri

Portfolio

Project-first, roughly reverse-chronological. The center of gravity is quantitative research — HPC, machine-learning factor models, and quantum computing — with a tail of builds that taught me how to ship. Each entry links to the code or a long-form write-up.

Research

RegimeFactorZoo — ML factor models with a regime-stability test · 2026 → present Fama-French and ML factor models on fully public data, testing whether sparse factor selections survive a volatility-regime change. → Read the write-up · Code

Quantum computing foundations — Hong Lab · 2026 → present The linear algebra under qubits, gates, entanglement, and measurement, and where quantum amplitude estimation beats classical Monte Carlo for pricing. → Read the write-up

qhpc_cache — cache behavior of finance kernels (BigRed200) · 2025 → 2026 Empirical L1-cache characterization of Cholesky, Monte Carlo, GARCH, and GEMM on AMD EPYC, measured with PAPI. Every claim traces to a hardware counter. → Read the write-up · Code

Systems and tooling

disk-archival-toolkit — bounded-memory filesystem audit · 2025 Streams multi-terabyte filesystem inventories to classify storage tiers and emit budget-aware archival manifests, all in bounded memory. → Read the write-up · Code

OpsPilot — local-first workflow agent YAML packs, memory, guarded execution, and a contracts-first design. → Code

Hackathons and applied AI

Captain Whiskers — autonomous on-chain trading agent · LabLab AI Trading Agents Variational-quantum portfolio optimization + BFT verification + post-quantum signatures in one agent. → Read the write-up · Code

legal-document-intelligence — agentic RAG for diligence LangGraph + Qdrant + vLLM (AMD MI300X) for property and M&A document review, built around retrieval faithfulness. → Read the write-up · Code

AI Ethics Coach — Chrome extension (MV3) Selected text through Gemini, action-oriented flags, selection bubble + modal + side panel. → Code

Writing

The Abstraction Dictionary — prompt-pattern book with figures and dual PDF exports. → Repo


Timeline — projects and certifications

A single chronological view. Projects in bold; certifications in italics.

2026

  • RegimeFactorZoo — ML factor-model pipeline begun; public-data rebuild after the WRDS summer-access pivot. (ongoing)
  • Quantum computing research — joined Prof. Yuxi Hong's lab (April); working through Sutor, Wong, and Nielsen-Chuang. (ongoing)
  • IBM Generative AI Engineering Professional Certificate — Coursera.
  • Captain Whiskers — LabLab AI Trading Agents hackathon: quantum portfolio optimization + BFT + post-quantum signatures.
  • legal-document-intelligence — agentic RAG diligence pipeline.
  • Head of Technology, IU Data Science Club — leading biweekly technical workshops.

2025

  • qhpc_cache — empirical cache-characterization study of finance kernels on BigRed200 (continued into 2026).
  • disk-archival-toolkit — bounded-memory filesystem audit and archival manifest generator.
  • AI Ethics Coach — MV3 Chrome extension shipped.
  • Fundamentals of Quantitative Modeling — Wharton Online (Coursera).
  • Financial Markets — Yale (Coursera).

Earlier

  • Kary & Mark — advertising venture past $10K revenue (Savannah, GA).
  • The Abstraction Dictionary — prompt-pattern book project.

Longer narratives live in the blog. Weekly motion is on the timeline.

The Abstraction Dictionary (Book)

**Summary:** A book-length project on prompt engineering and LLM interaction patterns—complete with generated cover art, experimental figures on word choice, and PDF exports in both compact and full layouts. **Problem or question that started it:** Most prompt advice is vibes in

AI Ethics Coach (Chrome Extension)

**Summary:** A Chrome MV3 extension that sends selected page text through Gemini 1.5 and returns plain-language flags for ethical and policy-flavored concerns—wrapped in UI people might actually use. **Problem or question that started it:** “AI ethics” often lives in PDFs nobody

Convo AI (Gamified Learning Platform)

**Summary:** Convo AI is a Streamlit app that teaches through a JSON-driven skill map, XP, streaks, badges, and gated milestones—so progression feels like a game without turning the content into trivia fluff. **Problem or question that started it:** Most “learn to prompt” sites

HPC Cache Research on BigRed200

**Summary:** I run full-budget 548-pricing workloads on IU’s BigRed200 cluster and turn raw runs into reusable evidence for reuse-distance, locality, and working-set arguments under Prof. Yuxi Hong. **Problem or question that started it:** Finance-oriented HPC codes stress memor

LOL News (React Native App)

**Summary:** A cross-platform mobile app that curates headlines by date, runs text-to-speech with a humor-aware filter, and keeps news sources swappable behind a clean service layer. **Problem or question that started it:** News apps either overwhelm you or flatten everything in

NCAA Final Four Analytics Challenge (February 2026)

**Summary:** I entered a public leaderboard competition on NCAA tournament outcomes, built models in Python on Zerve, and packaged the work into a presentation with a transcript so the reasoning survived the event. **Problem or question that started it:** Bracket intuition is fu

Claude Hackathon Telehealth Platform (March 2026)

**Summary:** Over a compressed hackathon window I shipped a telehealth-flavored portal that ties a knowledge graph, Twilio touchpoints, and Claude into one demo you could actually click through—not a slideware fantasy. **Problem or question that started it:** Telehealth products

Workflow Agent (OpsPilot)

**Summary:** OpsPilot is my local-first ops agent: workflow priors, observation, guarded execution, retrieval/memory, and an explicit boundary between on-device work and optional cloud. **Problem or question that started it:** Most assistants either hallucinate shell commands or

Full case files live in Website_Me/portfolio/ — edit markdown in Cursor, redeploy to publish.