Manila GMT+8 · open to work
I build production software with AI agents, and every project number on this page names the file it was read from.
666duplicate side effects with naive retry, 0 with sluice
16of 19 tables cleared on account deletion
0unreachable members, down from 8, and 25 of 59 never dispatched
230,059located hypocentres against 27 modelled slabs
0.0149AU worst blind-to-predicted gap disagreement
102of 102 Jepsen etcd verdicts matched
35 projects28 live30 measured
Every entry below is a thing that was measured, the file the number was read from, and the command that produces it again. If a claim here cannot show its working, it does not belong on the page.
Under fault injection, naive retry produced 666 duplicate side effects and sluice produced none, while refusing one call in five as indeterminate rather than guessing; the landing page serves those figures from the committed results file.
chaos/results/latest.jsongit clone https://github.com/jamessuuu/sluice && cd sluice && cat chaos/results/latest.jsonA privacy law that is not tested is a promise. Planting a token-shaped string in a gate answer proved it never reaches events.jsonl, and removing the redaction proved the test fails when it should.
tests/redaction.test.mjsgit clone https://github.com/jamessuuu/windlass && cd windlass && npm test
README command failing on a fresh install, missed by 51 green tests
proofpage's own README opened with a command that fails on a fresh install, and 51 passing tests never noticed, because every one of them ran inside the repo where the missing directory exists.
README.mdgit clone https://github.com/jamessuuu/proofpage && cd proofpage && git show cf40975 --stat
golden cases at 100% exact match, 0 enforced findings on its own CLAUDE.md, 0 measured rules shipped
Run on its own author's context file, chaff reports zero enforced findings, and that run is a committed eval rather than a demo; it ships zero measured rules today and says so in every line of output, because the evals that would admit one need paid model calls.
evals/cases/git clone https://github.com/jamessuuu/chaff && cd chaff && pnpm install && pnpm vitest run --project eval -t dogfood
Jepsen etcd verdicts matched
kedge's linearizability checker agrees with Porcupine's published verdict on every one of the 102 vendored Jepsen etcd histories that carry one, and it says so from a corpus command anyone can re-run rather than from a README sentence.
vendor/porcupine/jepsen/ (103 histories, MIT) replayed by src/cli.js corpusgit clone https://github.com/jamessuuu/kedge && cd kedge && npm ci && node src/cli.js corpus25 more are on the proof page, including the four where the tool caught its own author out.
I read my own mail and I answer it. Tell me what is broken or what you are trying to build, and if it is a job for somebody else the first reply will say so.
Manila, Philippines, GMT+8. Within 24 hours. jameslorenzsantos@gmail.com
If a number on this site does not reproduce, send me the command and the output, and I will fix the page.
The same content exists as a file tree, tabs and a real terminal, and every command you watch it run is one you can type.
The professional record, and where most of this site's visitors arrive from.
The source for the work catalogued here. Most of these repositories are public.
For employers hiring Filipino talent directly, with the verification that platform runs.
The Philippine job-market profile, for roles listed through it.
This is a portfolio you operate instead of scroll. You ask it something, it prints the answer, and it always shows the command it ran so nothing is hidden.
Seven agents built this site and each one answers for its own part of it. Press anything below and it will keep offering the next step.
| The cast | Seven real agents that built this site, one lane each. Ask one something outside its lane and it hands you to the right one by name. |
| The story | Five chapters, 2021 to now, from university to the current role. |
| The work | 35 projects, ranked by strength. The 6 leads come first, and every card carries a proof tier saying how far that one actually got. |
This is a portfolio you operate instead of scroll. You ask it something, it prints the answer, and it always shows the command it ran so nothing is hidden.
Seven agents built this site and each one answers for its own part of it. Press anything below and it will keep offering the next step.
The panel above is a workbench you can operate. Everything below is the same record as plain HTML, for a browser with JavaScript off or a crawler that does not run it.
sluice
Exactly-once side effects for agent tool calls, plus human approval gates that survive a process crash.
windlass
Pipelines as typed graphs with verifier edges and human gates. The runner has no model inside it, and exit 3 means stop and wait for a human.
kedge
A deterministic five-node Raft simulator with a linearizability checker, held to the published verdicts of 102 real Jepsen etcd histories.
flume
Event-time windowing with watermarks, and a checker for the result, measured on three real streams whose lateness spans four orders of magnitude.
cofferdam
Every crash point of a write-ahead log store visited under an explicit device model, then held against real process kills on NTFS and against node:sqlite.
driftwatch
A documentation-drift checker with exactly three outcomes, whose first real run across seven repositories reported ten failures and every one was a false positive.
proofpage
A receipts page that refuses to print a number it did not measure, and which caught its own README lying about its first command.
claude-code-boundary-guard
A PreToolUse hook that stops an agent in one project from writing into another: one file, no dependencies, and a self-test that has actually failed known-bad cases.
chaff
Static analysis for CLAUDE.md, AGENTS.md and tool definitions. Every enforced rule ships the eval that measured its effect.
dipmeter
Draws 230,059 located earthquakes at their real depth inside a transparent Earth against the 27 subduction surfaces USGS modelled, and separates out the 106,108 whose depth a locator assigned rather than solved.
orrery
Positions 1,562,531 catalogued small bodies every frame by solving Kepler's equation in a vertex shader from their own published orbital elements, and says on the page why the default view smears the gaps it is famous for.
snapgauge
Contract tests for MCP servers. Snapshot the schema and behavior, then fail CI when the next version moves.
provenote
A C2PA honesty test: drop an image, see its full provenance chain and exactly what that chain cannot prove, with nothing leaving your device.
dogwatch
A scheduled agent pipeline that publishes every run: cost, gate decisions, refusals, and the runs where it found nothing.
tiltmeter
Tells you when a new model release moves your agent harness off true. Every reading reproduces from a pinned commit.
shipgauge
A browser-ML shippability study: what a model actually costs to ship, measured in a real browser, with n printed.
assay
Brand kits as pure functions: one JSON config in, thirteen byte-identical artifacts out, and a contrast gate that refuses to generate below 3:1.
galley
Renders a repository's real commits, CI results and test counts into a release video, with every number on screen labeled by the source it came from.
halo-halo
A legible Taglish code-switching segmenter: every switch point shown, every label traced to a readable rule, down to the affix inside a single word.
swage
ASL fingerspelling handshape practice, graded by a classifier the project trained and evaluated itself, on a device that never sends the camera feed anywhere.
graticule
Paste your notes and get a map of how their wording relates, embedded entirely on-device, with the technique's own blind spot demonstrated on the page.
clarifier
Paste a CSV and let a physics simulation arrange it, then read a computed number that says whether the arrangement beat a plain scatter plot.
pitman
Shows exactly what a browser speech model heard on your speech, with per-word confidence and a diff against what you meant, entirely on-device. No accent scores.
orphanage
A crawl-graph auditor that found 40 problems on its first real site, and was wrong about all 40.
aeo-lab
A crawler-access scanner whose main feature is reporting what it cannot prove.
polis
Reads an agent ecosystem off disk, renders it as an isometric city you can walk around, and audits it while reading.
carillon
Your typing rhythm becomes sound in the browser, and a checker enforces that nothing is uploaded.
tally
An offline supply-chain and licensing auditor for npm projects, whose first real run flagged its own SPEC.md as a leaked test file.
consentry
A rules-based linter for US A2P 10DLC SMS campaign registrations that detects documented rejection causes offline and refuses to predict approval.
Klik
A hiring app where clients and freelancers match by swiping. Live on Cloudflare.
Business Operations Platform
An operations platform a real business runs on: purchase orders, stock takes, approvals.
AGENTJAMES Control
A desktop mission control for every project on one machine. Local only, and it still works with the network off.
ClubScope Insight Engine
A concept prototype where the language model is never allowed to produce a number: typed tools compute, every figure cites its evidence, and a verifier recomputes it before anything renders.
Kredit Hero: AI Credit App
Full stack work on an AI credit application, as a software engineer at Kredit Hero.
Agent James: Agent Console
This site: an IDE you operate, where every claim carries a receipt.