Full-stack engineer, 15 years across enterprise web, mobile, and cloud — now building production machine learning and LLM systems.
Most of my work lives in private org repositories. This account is where the open source is.
I'd wanted to publish open source for years and never had the time for it. So the lack of time became the experiment: can an LLM get me to a bar I'd actually sign my name to, without taking anything away from primary work?
It's still running, and it runs in the open. The code is public and it was built with an LLM — I'm not hiding that, and I'm not selling it either. AI slop is real. The variable isn't the tool, it's whether the approach is competent and whether the bar is mechanical: coverage floors, accessibility gates, payload budgets, and visual baselines run in CI. Generated code clears the same standard as hand-written or it doesn't merge. Nothing ships on my judgment being sharp that day.
The other half is that users don't grade your code. They grade whether the thing works for them. So the scoreboard is downloads — 75,000 to over 100,000 in roughly sixty days, all of it inside a year.
That's where the experiment stands.
JevLint-LE — the first cross-language linter for the questions people write for TypeSafe's Jev model, in the editor, on the command line and as an MCP server. Jev answers a badly worded question as fast and as confidently as a good one, so the mistake shows up later as wrong answers. This finds it in the code, before the request is sent.
It reads questions straight out of JSON, JavaScript, TypeScript, Python, Rust and Go, with no API key and no network calls. On the VS Code Marketplace and Open VSX, on npm as a command line for CI, and as an MCP server for agents.
The rules are measured, not asserted. I pulled questions from 1,411 public files across 720 repositories and hand-labelled 300 of them before the rules ran: about 1 in 20 had a clear wording defect. What the rules catch and miss on that sample is written down in the repo, misses included. A part of a question built at runtime is counted as not read, never guessed at.
letools.dev — 16 tools, all 16 published to Open VSX and the VS Code Marketplace, and all 16 on the official Model Context Protocol registry so agents get the same interface editors do. Counts come from the registries' own APIs, not from me.
Local, deterministic, exit codes as the API.
envsync-le |
spot missing keys across your .env files, with a markdown report |
secrets-le |
detect and sanitize credentials locally, before you commit |
regex-le |
find, test, and validate regular expressions with ReDoS screening |
scrape-le |
check whether a page is scrapeable before you write the scraper |
unicode-le |
find the Unicode that hides meaning: bidi controls, invisibles, homoglyphs, mixed scripts |
versions-le |
find where one dependency is constrained differently across a repository's manifests |
i18n-le |
identify the i18n library a project uses, then audit its catalogs by that library's rules |
string-le |
extract string values for i18n from JSON, YAML, CSV, TOML, INI, .env |
numbers-le |
extract numeric values from JSON, YAML, CSV, TOML, INI, .env |
dates-le |
extract and analyze dates from logs, configs, and code |
paths-le |
extract file paths from JS/TS imports, JSON, HTML, CSS, TOML, CSV, .env |
urls-le |
extract URLs from documentation, configs, and code |
colors-le |
extract and analyze colors from CSS, SCSS, LESS, Stylus, HTML, JS/TS, SVG |
units-le |
extract every quantity with its unit, normalized, and refuse the ambiguous ones by name |
ids-le |
extract every UUID, ULID, NanoID, ObjectId and Snowflake, and decode the time inside |
ips-le |
extract every IP address, CIDR block and MAC, normalized and classified by scope |
Two halves of one idea, in Rust, MIT: coordinates a computer-use agent can trust, because a human marked them.
pixelcoords freezes the screen, lets you mark regions with real shapes, and returns pixel-exact targets as versioned JSON — labeled crops, click code, verification with exit codes, self-healing relocation when the UI moves.
pixelactions is the execution half: click, type, chord, drag, scroll at those coordinates, then confirm the interaction landed. Chained CLI, flow files, or a line protocol any language can drive.
Built for driving desktop applications that never shipped an API — UI verification, accessibility auditing, and agent computer-use, where a guessed coordinate is a failed run. It cannot act on a coordinate a person did not verify first. That constraint is the point.
Fifteen years across automotive, finance, defense, healthcare, and agriculture — General Motors, JPMorgan Chase, L3Harris, T. Rowe Price, Nutrien Ag Solutions, RumbleOn, Integrated Auction Solutions, Brierley + Partners, Kofile Technologies.
Currently Lead AI Engineer at OffensiveEdge, building an end-to-end machine learning platform and the LLM product on top of it — training and serving, a ReAct agent harness with tool-calling and hybrid retrieval, and a public prediction ledger anchored to Bitcoin via OpenTimestamps and to the Sigstore Rekor transparency log. That work is private; the ledger and its verifier are not.
Earlier: lead developer on the first multilingual EMS communication platform — Texas Fire Chief's Lone Star Achievement Award, EMS World Top Innovation Award, a patent, acquired by ESO Solutions. And ADA / WCAG / Section 508 remediation at Fortune 500 scale, which is why axe runs on every page of my open source, in both themes, in CI.
Stack — Python · PyTorch · TypeScript · Rust · React · React Native · Next.js · Node.js · Bun · GraphQL · PostgreSQL · Redis · AWS · Docker · Kubernetes · Terraform · LLM · RAG · MCP · agentic systems · MLOps
nolindnaidoo.com · LinkedIn is the fastest way to reach me.