Hello, I'm

Dan Price

Senior Product Manager, Thomson Reuters

I went from writing musical counterpoint to advocating in court to building AI-first products. Today I'm building Westlaw's next-generation research experiences so attorneys can trust what the machines deliver.

Musician Lawyer Vibe Coder Founder Sr. PM

See experience Get in touch

Latest post: Evolving a DSPy Program to Solve ARC-AGI-2

Experience

Professional Journey

From legal practice to legal technology innovation

Senior Product Manager

Thomson Reuters

May 2025 – Present

Building new AI features for Westlaw, informed by what I built and learned at my startup, LexMagic.

Design and prototype research tools.

Work with design, engineering, and labs to ship.
Founder & Developer

LexMagic

Jan 2024 – May 2025

Built a research tool for Louisiana lawyers.

Answer plain-English questions with cited cases.

Collected and organized Louisiana caselaw and statutes.

Built end-to-end and spoke with every user.

Used daily by several Louisiana firms.

Shut it down to bring my experience to Westlaw at scale.
Associate Attorney

Roedel Parsons Blache Fontana Piontek & Pisano, APLC

Oct 2020 – May 2025

Litigation attorney; taught myself to code to work faster.

Handled business disputes in court

Built tools to automate research

Taught myself programming and AI while practicing law

Projects

Project Deep Dives

Side projects and experiments in AI, music, and code

01 Prime Directive View details 02 Dan in Japan View details 03 LLM to MIDI View details 04 LexMagic View details 05 Hands-Free Claude Code View details

Project Details

Prime Directive

ai claude-code autonomous-agent experiment

An experiment where I gave Claude Code a single goal ('start a business and earn $100') and let it run autonomously. The agent built products, sent 1,400+ outreach emails, submitted to 50+ directories, and even applied to Y Combinator.

Supervisor loop keeps the agent working until success criteria are met or it explicitly asks for help.
Progressive memory system survives context resets: CLAUDE.md hub + domain-specific memory files.
Discord integration for human-in-the-loop: agent can DM me when it's stuck, I can reply anytime.
Built 9 products autonomously including accessibility audits, LinkedIn post generators, and logo makers.
Got real engagement: Kinde acknowledged a bug the agent found; Kahoot expressed interest.
Agent submitted a YC application on its own initiative.

Visit the Shop

Dan in Japan

ai claude cursor photography

AI-powered photo gallery documenting a 2-week trip to Tokyo, Osaka, and Kyoto with nearly 2000 photos, auto-captioned using Claude 3.5 Sonnet and SerpAPI.

Built complete website using Cursor IDE with AI pair programming.
Auto-generated captions using Claude 3.5 Sonnet with photo metadata and SerpAPI for location context.
Used Instructor library to extract cultural and historical tidbits from each image.
Interactive gallery with filtering by location and chronological sorting.
100% of captions and points of interest are AI-generated from dynamically generated LLM prompt context.

View Live Site GitHub

LLM to MIDI

music llm midi baml

Procedurally generate musical compositions using LLMs through a structured, modular approach with multiple voices (bass, tenor, alto, soprano, piano).

Uses BAML to structure LLM prompts for music generation with dynamic model selection.
Generates multi-track MIDI files with bass, tenor, alto, soprano, piano, and percussion.
Two-step process: creates composition plan, then generates musical content.
Outputs MIDI files and JSON logs capturing composition plan and final piece data.

GitHub Repo

5-Voice Composition Audio

Bittersweet Carnival Waltz - A Minor Audio

5-Voice Ensemble Audio

LexMagic

legal-tech ai founder

AI legal research startup built for Louisiana attorneys with citation-backed answers.

Ingested Louisiana caselaw/statutes to support neural retrieval.
Produced plain-English responses with pinned citations.
Used daily by several Louisiana law firms.
Built end-to-end and spoke with every user.
Shut down to bring experience to Westlaw at scale.

Hands-Free Claude Code

voice ai python macos

Voice-first interface for Claude Code with a floating HUD, VAD-only capture, local STT, and spoken replies. One CLI boots everything; preferences from a single config file.

Hands-free loop: VAD capture → faster-whisper STT → Claude Code (streaming) → HUD → TTS → back to listening.
Floating HUD (CustomTkinter) with mic mute and screenshot toggles, plus optional typed input.
Config-only setup: single config.toml drives all behavior; state files live under platformdirs directory.
WebRTC VAD-gated capture with optional guided calibration and sensitivity slider.
Kokoro ONNX TTS for high-quality spoken responses or macOS 'say' fallback.
Optional per-turn screenshots saved and attached for visual context.