Software engineer. Independent builder.

Tools

• OpenAI Agents SDK contribution

  Fixed a circular dependency in the Python voice-streaming example.

• Live Transcriber

  Real-time microphone transcription with Deepgram.

  GitHub · PyPI

  15K+ downloads · PePy

• indentlogger

  Configurable nested-call tracing and automatic instrumentation.

  GitHub · PyPI · Downloads on PePy

• Public Health Communication Reviewer

  Health wording analysis for human review.

  GitHub

• Meeting Canvas

  Draw and connect components, edit labels nearby, and save diagrams in your browser.

  Open Meeting Canvas →

• Spatial Lab

  Explore a synthetic room with a two-link robot arm. Adjust joints, place the base and export the plan.

  Open Spatial Lab →

• BugSketch

  Describe an issue or start with a screenshot, then review and copy or download locally.

  Describe → Optional evidence → ReviewOpen-source widget · no account · local export

• System Design Studio

  Work through a URL-shortener design: requirements, API, traffic assumptions and a reasoned tradeoff. Save your attempt locally.

  Requirements → Interface → TradeoffsExplore reads, writes and cache assumptions →

• AI-Assisted Health Services Navigation

  Working with Professor Arunim Garg on AI-assisted decision-making research to help older adults navigate health services.

• Bob

  Personal voice assistant for hands-free development with local Whisper transcription.

• Vision Lab

  Robot-path animation and visibility checks.

• Food Diary

  Log portions and manage a food diary in your browser.

  Open Food Diary →

• Small Studio

  Edit source, preview it in isolation and save browser-local revisions. Import and export JSON packages.

  Open the editor →

• Pick a coding model

  Completion cost depends on priced input, output and reasoning tokens, plus repeat attempts. Compare score with reported cost per benchmark attempt to see the tradeoff.

  Budget · Luna launch chartGPT-6 Lunamax66.59%$0.2169 / attemptHigher score · Sol/Astra launch chartGPT-6.1 Solmedium73.01%$0.4196 / attempt

  DeepSWE v1.1 · separate OpenAI launch charts, September 22 and 29, 2026. Snapshot: October 1, 2026.

  Sources & selection

  Each pick is the lowest reported cost within 3 score points of its own chart’s best. Luna and Sol/Astra use separate launch charts; these are source-specific starting picks, not a matched head-to-head ranking.

  Costs are publisher-reported averages per attempt. Token breakdowns, retries, budgets and final billed totals were not reported. Datacurve’s reported averages and launch-rate estimates stay separate in the detail view; its pricing-group picks use 10- and 3-point thresholds.

  Cost per successful completion needs failed attempts and retries in the total. Cost-per-task methodology · Cost-aware agent evaluation

  Luna chart · Sep 22 · Sol/Astra chart · Sep 29 · Full graph, tables & sources →