Edwin Jorge — Senior Software Engineer

I build software
people and agents rely on.

Senior Software Engineer at Mercado Libre, shipping full-stack web platforms in production at LATAM e-commerce scale (4.3M+ users). Outside work, I keep building — including a set of real, publicly running AI-agent projects: webhook-triggered bots and a live MCP server you can question right now.

Ask an agent about me ↓See the work →

01 · Ask about my work, live

Ask it something. Right here.

This box calls the same live ask_about_edgeorgie engine any MCP client would — real TF-IDF retrieval over my résumé, case studies, and reliability reports, cited by file and line.

Ask a question to see the live answerMode — generated prose if an LLM key is active on this deployment, or a direct cited excerpt dump if not. Retrieval and citations are real either way.

ask-edgeorgie-mcp · live · streamable-httpconnected

Ask a real question below — e.g. “What has this person built with AI agents?”

❯

Prefer to connect your own agent?

{
  "mcpServers": {
    "ask-edgeorgie": {
      "url": "https://ask-edgeorgie-mcp.vercel.app/mcp"
    }
  }
}

Drop this into Claude Desktop or Cursor for get_experience · get_projects · ask_about_edgeorgie. Same engine as the box above. Open the standalone endpoint ↗

02 · Shipped

Four real, running projects — not slideware.

A deployed MCP server you can query about me, a webhook-triggered bot, a CI gate, and a publicly reachable repo Q&A server — each with a GitHub run log behind its numbers.

01

ask-edgeorgie-mcp

Don't take my word for it — ask an agent directly.

A live MCP server about ME, not a repo. get_experience / get_projects / ask_about_edgeorgie, backed by TF-IDF retrieval with exact file+line citations over my real resume, case studies, and reliability reports. Synthesizes grounded prose with a real model (Claude Haiku) when an LLM key is configured, citing every claim back to a source excerpt — falls back to a deterministic citation dump otherwise, same corpus either way, no invented answers. Scoped and guarded: it only answers questions about me, declines off-topic requests, and resists prompt injection from its own retrieved text. Verified by a real @modelcontextprotocol/sdk client over Streamable HTTP against the live deployed URL, not a local stdio demo.

TypeScriptMCP SDKVercelTF-IDF retrieval
GitHub ↗Ask me live ↗
Tools exposed over MCP
get_experience()get_projects()ask_about_edgeorgie()
try it in the box above ↑
Live, public, no auth wallStatus
real external MCP client (HTTP)Verified by
5/5 passingLocal tests
Streamable HTTP + stdioTransport
02

triage-desk

A GitHub bot that triages issues on its own — no human click.

Webhook-triggered agent bot that reads a new GitHub issue, classifies kind/priority, flags likely duplicates, posts a comment, and applies labels — triggered by GitHub's own `issues: opened` webhook, not a UI button. Falls back to a deterministic heuristic when no LLM key is configured, so the full pipeline produces real, non-fabricated output end-to-end either way.

TypeScriptGitHub ActionsOctokitNode.js
GitHub ↗Live Actions run ↗PR #25 (webhook bot) ↗
Live example · webhook payload → label

issues: opened

Bug: duplicate detection crashes on empty issue body

Opening an issue with no body text throws inside the dedup step before labels get applied.

↳ classified:bugp0confidence 40%
real run · issue #24 ↗
22s, successReal webhook run
83.3% (15/18)Kind accuracy
83.3% (15/18)Priority accuracy
22/22 passingLocal tests
03

eval-lab

An installable CLI + GitHub Action that fails CI on a bad prompt.

Variant runner, deterministic checks, LLM judge, and regression diff — packaged as a CLI (`npx eval-lab run`) and a reusable composite GitHub Action. Dogfooded on its own CI and on triage-desk's CI: it ran triage-desk's real heuristic-triage function directly and gated a real PR on the result.

Node.js ESMGitHub ActionsCLI
GitHub ↗Benchmark run ↗Dogfooded on triage-desk CI ↗
24/24 cells passing100%

point-in-time result from the linked CI run, not a live feed

benchmark run · 3 passes ↗
100% (24/24 cells, 3 runs)Benchmark pass rate
~20ms / cellAvg latency
$0 (offline demo model)Cost
7/7 passingLocal tests
04

repoask-mcp

A live MCP server, publicly deployed, answering real questions about real repos.

An MCP server built on the official @modelcontextprotocol/sdk exposing index_repo / ask_repo / list_indexed_repos tools over Streamable HTTP. It's deployed and publicly reachable right now — a real external MCP client connected to it over the open internet and got real citations back, not a local demo.

TypeScriptMCP SDKVercelExpress
GitHub ↗Live MCP endpoint ↗
Live, public, no auth wallStatus
real external MCP client (HTTP)Verified by
3/3 passingLocal tests
Streamable HTTP + stdioTransport
05

crispy-profiling

React re-render profiler, published to npm, for humans and agents alike.

A React re-render profiler shipped as four surfaces off one engine: CLI, GitHub Action, MCP server, and an Agent Skill so Claude Code, Cursor, Codex and similar tools can measure a re-render fix instead of guessing at one. Opens the target app in headless Chromium, drives described interactions, and reports which components re-rendered, how many times, and why (props/state/context/parent) — validated against 5 real open-source apps (Redux Essentials, Next.js App Router Playground, Excalidraw, shadcn-admin, react-admin), finding a fixable re-render problem in each.

TypeScriptPlaywright/ChromiumMCP SDKGitHub Actions
GitHub ↗npm ↗
Published on npmStatus
5 real OSS appsValidated against
19 tested, 18.3/19.0 validatedReact versions
CLI + Action + MCP + Agent SkillSurfaces
06

simplescope

Learn algorithms by watching real JavaScript run, not static diagrams.

An interactive, visual, gamified platform for learning algorithms — step through real executing JavaScript instead of reading pseudo-code. Built with Next.js App Router, TypeScript, Tailwind v4, and MDX lessons; fully functional with zero required environment variables (progress falls back to browser localStorage when optional account sync isn't configured).

TypeScriptNext.jsTailwind v4shadcn/uiMDX
GitHub ↗Live site ↗
Live, publicly deployedStatus
0 — works out of the boxRequired env vars
Next.js App Router + TypeScriptStack

03 · Experience

Production engineering at LATAM scale.

Mercado Libre

November 2022 — Present

Remote

Senior Software Engineer

  • Owned 10+ initiatives end-to-end solo, together reaching 4M+ users across LATAM; one shipped feature alone served 1.5M+ users with zero production incidents.
  • Led cross-platform initiatives spanning web, mobile WebViews, and AI-assisted native development in Kotlin (Android) and Swift (iOS), using Spec-Driven Development and scalable frontend architecture.
  • Built backend APIs with caching, idempotency, and low-latency design underpinning those same web, mobile, and TV surfaces.
  • Shipped smart TV and OTT surfaces: web receivers for LG, Samsung (Tizen), and a Chromecast Web Receiver, for video streaming.
  • Implemented observability and monitoring using Kibana, New Relic, Datadog, and Grafana aligned with SRE practices (SLA/SLI/SLO).
  • Applied Clean Architecture and Domain-Driven Design while refactoring legacy systems and improving code quality.
  • Designed and owned AI-agent architectures end-to-end — defining agent rules, skills, and cross-service orchestration — as a sustained, multi-year production responsibility since Nov 2022, not a one-off project.

Vansa

February 2022 — July 2022

Remote

Front-end Developer

Built React frontend applications and responsive dashboards; applied Atomic Design and Clean Architecture.

Freelance

April 2020 — February 2022

Remote

Front-end Developer & UI/UX Designer

Designed frontend architecture/design systems, translated Figma designs into scalable React interfaces.

05 · Beyond the code

The same habit, from age 15 to now.

Build it yourself to understand it, whatever “it” is. Tap any step for the real story.

04 · How I use AI

How I use AI on a daily basis.

I treat it as an orchestration layer with its own operating discipline, built the same way I'd build any system I'm accountable for in production.

Live from the GitHub API — not a static mockup

fetching live repo data…

04 · Contact

Want to see more of the work?

Everything above links to a real repo, run, or deployment — no invented metrics. Happy to walk through any of it.

ed.jorge1122@gmail.comgithub.com/edgeorgie ↗