I build systems that turn complicated work into something people can actually operate—from ETL and realtime document agents to private memory, interview practice, physical-AI data, and reflective tools.
Both remove an expensive layer of friction: one around AI artifacts, the other around production data pipelines.
A place to publish, edit, and collaborate on AI-generated HTML. It turns the thing an agent made into a live artifact with comments, version history, real-time collaboration, and a path into narrated explainers and courses.
Visit HTML Docs ↗
A fully managed ETL platform and drop-in Fivetran replacement. Skybright handles connectors, white-glove migration, hosted pipelines, monitoring, maintenance, and day-to-day operations—while keeping the path to an in-house stack documented.
Visit Skybright ↗
Preparation, privacy, reflection, and physical data all get better when the evidence and the next action live in the same place.
One preparation engine with all-company, OpenAI, and Anthropic editions. A researched question bank opens directly into coding, PM response, and system-design workspaces with validation and a context-aware coach.
A digital-footprint command center that discovers exposure, confirms identity matches, routes each source to an honest removal path, and keeps proof for later rechecks.
A philosophy-guided journal for web and mobile, with voice reflection, named philosophical lenses, searchable entries, and a memory ledger the user can inspect and control.
A physical capability foundry that starts with one robot task below its deployment bar, then connects real episodes, calibrated simulation, targeted synthetic variants, and held-out evaluation into one acquisition program.
These projects explore a narrower behavior: making a document agent safe, turning conversation archives into insight, following a model portfolio, or helping someone reflect on their own history.
Custom, consented physical-audio collection designed around a model capability gap.
A realtime voice partner that reads and safely edits Google Docs and HTML Docs.
Model-portfolio mirroring, rebalance actions, trade alerts, and notification routing. GitHub ↗
Private ZIP import, conversation reading, and structured communication analysis. GitHub ↗
An iOS experiment for multi-source conversation imports, on-device sentiment, profiles, and friend graphs.
A SwiftUI medication and wellness tracker for shots, symptoms, voice logs, and HealthKit signals.
An iOS reflection companion combining AI conversation, journal history, and user-owned storage.
Turns an AI conversation into a reviewable first-person journal entry and reflection history. GitHub ↗
I build source-grounded courses and labs around a demonstrated outcome. The learner leaves with a design, benchmark, capstone, or tested implementation—not a pile of notes.
From policy loops and demonstrations to VLAs, serving, evaluation, and a measured capstone.
Seven chapters on latent dynamics, imagination-based planning, uncertainty, and embodied systems.
Choose a workload wedge, ship the serving MVP, benchmark the promise, and earn the hardware path.
A deterministic six-hour lab covering estimation, noisy data, fake hardware, public tests, and a hidden-style sandbox.
HTML Docs — first-party product page, retrieved July 2026
Skybright — first-party product page, retrieved July 2026
Interview Loop — local product specification and product UI capture
Clean Slate — local product specification and product UI capture
Saarthi — local product specification and App Store capture
World Data — local product specification and landing-page capture
World Audio — local product specification
Rubber Duck — repository readme
TradeTwin — repository readme
WhatsApp Archive — product source
Vibely — native app source
Dosey — native app source
Chia — repository readme
Chatter Scribe — product source
Robot Learning, Gently — course map
World Models — course map
Inference Company Founder — course map
Servo Velocity Lab — repository readme
Inventory frozen July 28, 2026. Included: 18 substantive owned builds with local source evidence or a first-party product page. Excluded: empty repositories, interview-note and resume workspaces, generated output, credentials, raw private data, and third-party checkouts.