SPEC → IMPLEMENT → VERIFY → SHIP

From Spec to production. Built with agentic workflows.

I'm Javier Ramos, a backend-leaning full-stack engineer with 4+ years shipping production systems. Everything here is real: deployed products, live GitHub activity, and the agent-driven process that built each one, from spec to green CI.

SANTIAGO, CHILE (GMT-4) · REMOTE · EN / ES

View the workSee the method
In a hurry? Read the CV
example claude code agentic workflow
/prd portfolio-exploration.md
plan PRD → 9 tasks · acceptance set
tests 7 specs written first · red
/implement portfolio-prd.md
iterate Claude Code → diff applied
/verify → 7/7 tests passed
lighthouse 96-100 · live site
ship deployed · main@a1f9c2
/learnings → 1 skill improved, 2 repo rules modified, and 1 memory entry added
Contributions · 12 moREAL
2,528

verified on GitHub

Ship time · this siteREAL
<24h

spec to production

Lighthouse · live siteREAL
96-100

every category, desktop + mobile

Production experienceREAL
4+ yrs

payments, e-commerce, biotech

FEATURED EXHIBITS

The product and how it was built

View all exhibits

Each exhibit pairs the shipped product with the build behind it: the PRD it started from, the timeline, and the metrics, honestly tagged.

  • WEB_APPPublic repo
    This Portfolio

    The showroom itself: a full-stack Next.js + NestJS app taken from spec to production in under 24 hours of agent-driven work.

    Built with

    Claude CodeMCPPRD loopNext.jsNestJSPostgreSQLPrisma
    Live Repo
    Ship time <24hREAL
  • WEB_APPPublic repo
    Portfolio Agent

    This site's own concierge: a retrieval-augmented agent that answers questions about Javier, grounded in a curated corpus and live GitHub activity, with a published eval to back it.

    Built with

    Claude CodeMCPPRD loopNext.jsLangGraph.jspgvectorOpenRouterGeminiGroqLangfuseUpstash Redis
    Live Repo
    Eval accuracy 100%REAL
  • CASE_STUDY Private code
    territory-developer

    A geography-first isometric city-builder, a year in the making, driven by a custom Unity-to-agent bridge. Public writeup, private code.

    Built with

    Claude Code (CLI + desktop)CursorCustom unity-bridge MCPPRD / implement / verify skillsUnity 2022C#Next.jsPostgreSQLMCP
    Stage Beta MVPREAL
  • $ blipsmith render blip.spec.json
✓ blipsmith-out/blip.wav
→ 220 ms · peak -1.0 dBFS · f0 653 Hz
✓ 35 goldens match · linux + macos
    TOOLINGPublic repo
    blipsmith

    An agent-playable lo-fi synthesizer. There is no model inside: an LLM turns the knobs over MCP and a deterministic C++ core renders the bytes, identical on every platform.

    Built with

    Claude Codeprd-grill-mesequential backlogC++17Pythonpybind11MCPGitHub Actions
    Repo
    Bit-identical 2 platformsREAL
  • $ agent run --kit
✓ skills · prd · implement · verify
→ mcp · schema · scaffold · deploy
✓ ready
    TOOLINGPublic repo
    agentic-dev-kit

    The open-source toolkit behind every exhibit: Claude Code skills, an MCP server, and PRD templates that make agentic building repeatable.

    Built with

    CC skillsMCP serverPRD templatesTypeScriptNode.jsMCP
    Repo
    CI greenTARGET
  • WEB_APPPublic repo
    World Music Map

    Explore the planet by sound: an interactive map where each curated place opens region-defining music, with full attribution built in.

    Built with

    Claude CodeMCPprd-grill-mePRD loopNext.jsTypeScriptPrismaPostgreSQLMapLibre GL
    Live Repo
    Stage S1 deployedREAL

THE METHOD

A closed agentic loop, on every exhibit

Each product is built with a repeatable, verifiable pipeline, driven by a Spec and Claude Code skills, closed by automated checks.

01

Explore

Scan the repo, read context, frame the problem with the agent.

02

PRD

Write a PRD: scope, tasks, and acceptance criteria. It is the spec the build runs against.

03

Grill

Pressure-test the PRD: polling rounds lock design, behavior, and dependencies.

04

Implement

The agent writes code and tests against the PRD.

05

Verify

Tests from the spec, plus lint, typecheck, and CI, gate every ship.

06

Learn

Reflect on session process and results and improve the repo context, memory, and skills.

iterate When verify fails, loop back to implement. Learn then feeds each session's lessons back into the repo context, memory, and skills, and Claude Design ⇄ UI runs on every visual change.

PROOF · GITHUB ACTIVITY

Backed by real, continuous work

2,528 contributions · last 12 months · verified

Contribution heatmap

JulAugSepOctNovDecJanFebMarAprMayJun
LessMore

Primary stacks

the languages & frameworks I build in

  • TypeScript
  • JavaScript
  • C#
  • Python
  • SQL

Contributions over time

GitHub commits + PRs, contrib / week

3441720

2,528 contributions over 1Y · peak 344 per week

CONTACT

Let's build something

Hiring full-time? Senior full-stack, backend-leaning, remote-first. The CV has the facts; the exhibits have the proof.

Read the CV

Have a project? Freelance work here means scoped full-stack products, APIs, and agentic-workflow enablement for teams that want their engineers shipping with agents. Tell me what you need.

jiramos87@gmail.com