Prasith Govin

Founder, CTO, agent builder

I build small teams of AI agents that ship real software.

Prasith Govin photoshopped as Steve Jobs, in a black turtleneck holding an iPhone in front of the Apple logo at a keynote.

CTO and founder. Years running enterprise engineering, now building lean teams where a few people plus a fleet of agents do the work of a much larger org.

Tampa, FL GitHub X LinkedIn

01Building

4 active

  1. Parker

    github.com/prasithg/parker

    A voice-first assistant for people with effortful speech, built for my dad, who has Parkinson's. Real-audio evals improved comprehension from 58% to 82% with repair. Live sessions now leave a local review timeline. 1,124 tests.

    Watch the 30-second teaser
    Research prototype. Synthetic demo. Not medical advice.
    On-screen text
    1. Live caption: remind… mom… med… no, message… tomorrow
    2. Most assistants guess.
    3. A generic assistant card, "Sending medication message to Mom now.", struck through.
    4. Parker asks.
    5. Parker card. I heard: remind mom med message tomorrow. Did you mean: A, create a reminder for tomorrow. B, draft a message to Mom. C, neither, try again. B is picked.
    6. Draft only. Nothing sent. Stays on the machine until a human reviews it.
    7. You stay in control, next to the real caregiver review screen.
    8. Built for effortful speech. Repair, confirm, stage, review.
    9. Parker. Research prototype. Synthetic demo. Not medical advice.

  2. Clawrari

    clawrari.com

    My agent-orchestration stack, open-sourced: memory architecture, meta-learning loops, playbooks that let one operator direct many agents. v0.5.0, 8 production skills.

  3. trace-to-tripwire

    github.com/prasithg/trace-to-tripwire

    Metadata-only receipts and fail-closed tripwires for agent trajectories. No content capture, zero deps, CI-gated.

  4. JobLeap / WorkConnect

    jobleap.ai

    Shift marketplace and orchestration for healthcare staffing: matching, credentialing, scheduling, pay. Built with UKG Labs as partner and investor.

02How I build

  1. planPlan before code. The design doc comes first. If I can't write it down, it isn't ready to build.
  2. buildAgents build, I direct. Coding agents write the code. I set the spec, review the diff, and own the decision.
  3. reviewCross-model review by default. Every change is critiqued by a different model family before it ships.
  4. measureMeasured numbers only. Evals run with negative controls. I report what's measured, never what I'm hoping for.
  5. guardFailures become guardrails. Each failure turns into an executable check, not an apology.
  6. queueA night queue. Agents build while I sleep, and the morning is for review.

03Build log

Receipts only
  1. Sep 2026Parker teaser, 30 seconds, rendered in code from my own brief: light theme, my copy, the model’s motionreviewnote →
  2. Sep 2026Memory health check now exits 3 when it can’t run, instead of freezing a healthy system with a false REDreviewnote →
  3. Sep 2026This site got a shell contract: CI fails if the hero, portrait, nav, or CSS change outside an explicit redesigngithub ↗
  4. Sep 2026Parker: “yes one” now selects choice one, a fix from the first human-tester sessiongithub ↗
  5. Sep 2026Clawrari: 26 distilled playbooks in 29 commits since mid-August, from eval evidence to retry boundariesgithub ↗
  6. Aug 2026Parker scenario gauntlet: 73 new scenarios over three rounds, 13 fixes, live-verifiedgithub ↗
  7. Aug 2026Parker fast-voice orchestrator: lookups and actions run in the background while the conversation keeps goinggithub ↗
  8. Aug 2026Added Parker's local session-review timeline: turns, guard trips, lookup latency, staged actions, and one-tap feedback tied to the exact eventgithub ↗
  9. Jul 2026Shipped clawrari.com: brand, landing, and story pages, three review passes across two model familiesclawrari.com ↗
  10. Jul 2026Parker real-audio eval on dysarthria corpora: 58% → 82% comprehension with repair loopnote →
  11. Jul 2026trace-to-tripwire v0.1: fail-closed CI gate for agent trajectory receiptsgithub ↗
  12. Jul 2026Two agents (Claw + Hermes) now build and review each other's work over a private channelnote →
  13. Jun 2026Cross-agent eval framework: strict scorecards with negative controlsgithub ↗

04Notes

13 entries All notes RSS
  1. Sep 2026 a check that couldn't run said red My memory health check froze a healthy system after an upgrade, and fixing that false alarm is what made the next, real RED believable.
  2. Sep 2026 picking the colors before the model does The Opus product videos I saw all looked alike, so for the Parker teaser I wrote the palette, the copy, and a banned list before the model wrote any code.
  3. Sep 2026 the visual test passed and the page still looked wrong My site passed its first visual review because the checks measured fit, not the composition I had approved.
  4. Jul 2026 my own repo disproved my readme I caught my own README overclaiming Parker, and the real eval number was better than the lie.
  5. Jul 2026 a repo can commit daily and still rot My agent audited our open-source sync and found three months of drift behind green CI and a daily commit streak.
  6. Jul 2026 guardrails, not prompts A prompt tweak fixes one bad output; a guardrail in the loop stops every one that fails the same way.
  7. Jul 2026 managing coding agents is managing juniors who forget everything overnight Coding agents are capable juniors with no overnight memory, so the work is in the brief, not the model.
  8. Jul 2026 a subagent dies quietly: the context-death contract Five subagents died in two weeks, all context blowouts, so the anti-death rules became a required contract on every spawn.
  9. Jul 2026 metadata-only receipts: trace, then tripwire Every agent run leaves a metadata-only trace that a fail-closed tripwire gates on before the next run proceeds.
  10. Jun 2026 the night build queue I set the plan at 10:30pm and agents build while I sleep; it only works if every run re-derives its numbers by morning.
  11. Jun 2026 two agents reviewing each other over a private channel Two agents on two machines reviewing each other produced our most novel work, and a six-hour deadlock that taught the most.
  12. May 2026 cross-model review is the default Every change gets reviewed by a different model family before it ships: one writes, one reviews, single round.
  13. May 2026 memory is the product surface For an agent, memory isn't a bolt-on feature; it gets a measured architecture with benchmarked recall and negative controls.