schultz.ai
Ask my AI

§ 08 Changelog

What I’ve shipped lately.

The AI agent and this site are a living artifact — every feature, fix, and polish that lands shows up here first. Newest at the top, dates on the rail.

13 entries·11 new features
  1. New

    Screen Test — a grounded talking-head video of me

    Name a topic and I deliver a short take to camera — in my own voice. The brain writes a tight ~8-second script true to the record (self-scored), my cloned voice reads it, and a video model animates a reference photo of me to deliver it, lip-synced to my actual voice. A grounded synthetic clip, clearly labelled AI-generated. You read the script while it renders, then watch the take. Daily-capped, with a clear 'not configured' note until the keys are set. Also: the app store now has search, filters, and sort.

  2. New

    Frontier Watch & Deep Dive — research, two ways

    Frontier Watch — a live integration: pick a frontier-AI topic and the brain pulls the leading arXiv research (no key) and maps each paper onto my real record — shipped it, adjacent, or ahead of where I've been — honestly and with the receipts (new SDK recipe frontier()). Deep Dive — a research agent that shows its work: ask one hard question and watch it plan the lines of inquiry, research each against the record live, then synthesize one grounded answer and self-score its own groundedness before it lands (new SDK recipes plan() + synthesize(), composed with ask()).

  3. New

    Time Machine & Live Roles — two new Lab apps

    Time Machine — drag or play through the years and watch my career unfold: the voice → multimodal → ambient arc lights up, the active role highlights, and a card surfaces the real numbers at every point (instant, client-side, no model). Live Roles — a real 3rd-party integration: it pulls this month's 'Who is hiring?' postings live from Hacker News (no key), then the brain ranks the senior AI/product/leadership openings by honest fit to my record — a fit score, the cited why, and the single biggest gap for each (new brain-SDK recipe shortlist()).

  4. New

    The Anthem — a real AI song from my brain

    Name a topic and pick a vibe, and the brain writes a song about my career — every lyric grounded in my actual record (new brain-SDK recipe song()) — then records it into a real track via ElevenLabs Music that sings the grounded lyrics. Daily-capped (music gen is pricey) and it degrades to a lyric sheet if music isn't enabled. Also: the SDK now separates true primitives (ask/decide/judge/verify) from recipes (the app-shaped moves like brief/remix/song).

  5. New

    The Briefing Room & Remix — new Lab apps

    The Briefing Room — pick a lens and the brain scripts a two-voice 'deep dive' on my record (a curious host + a skeptic who pushes on the gaps) and reads it back in two real voices, highlighting each cited line. Remix — ask one question and get four honest, cited framings, each self-scored, with upscale-to-full-argument. Each ships its own app-eval, built on new brain-SDK calls (brief, remix).

  6. New

    Three new Lab apps built on the brain

    Cold Open — drop a role or company and get a grounded fit read plus the short, cited outreach I'd actually send (built on fit()). Two Truths & a Lie — a game on the anti-hallucination spine: the brain deals two real, cited career facts and one plausible fake it made up on purpose; spot the lie, then see the receipts. Career Constellation — the whole record as one navigable graph: pick a skill and watch it draw a constellation across the roles it spans, with the real bullets underneath (instant, client-side, no model). All three live in the gallery at /lab.

  7. New

    Studio + the brain SDK — build your own app on my brain

    Labs is now a developer platform. A real SDK exposes the brain as functions — corpus(), roles(), search(), grounding() (instant, from the record) and ask(), decide(), fit(), judge() (live, model-backed, grounded). The new Studio app is a sandboxed code editor: write JS against the SDK, run it live, get console output — hello-world apps, grounded questions, opinion calls, fit checks, and your own evals — with the full API reference built in.

  8. Release

    schultz.ai Labs — a personal AI platform

    Every app here is built on my data — what I've done, and how my AI performs. New: The Judgment Call (/lab/judgment-call) takes a real consumer-AI decision and returns a grounded, opinionated take cited to my actual record, honest where it runs out, and self-scored for grounding before it speaks — shipped only after its four app-evals passed. The hiring panel, live evals, and insights now live in the gallery at /lab (old links redirect).

  9. New

    Streaming voice mode — the AI speaks in my cloned voice

    Click the speaker icon in the chat header to hear responses in my voice (ElevenLabs). Streaming chunks start playing within ~1s. Mic button in the input dictates your questions via the browser's speech API.

  10. New

    Reverse interview mode

    Click "Interview me instead" in the chat — my AI flips the dynamic and interviews you about the role / opportunity, then synthesizes a fit read from my POV.

  11. New

    Import me into your AI (MCP server)

    Connect Claude Desktop, Cursor, or any MCP-compatible client to my site. Your AI can call my resume + tools directly. Setup at /import.

  12. New

    Tailored one-page PDF generation

    Ask the AI for a one-pager tailored to your role — it generates a real letter-size PDF with the highlights reframed for your audience.

  13. Persistent share links with AI-generated titles

    Hit "share" on any chat — get a short URL with a contextual title (e.g. "Robert's Meta wearables RLAIF work"). Backed by Vercel Blob; survives forever.