# Manuel Jiménez

> I put AI agents into production.

AI Product Manager at Amazon: I lead AVS 2.0, an AI operating model that changed how 80 people work and aligned 5 countries. After hours I build my own AI apps: a social app with 1,400+ users and a running coach that is an AI agent.

## Projects

### AVS 2.0 · Amazon — AI does the work. A human decides.

The AI platform with agentic skills for Amazon Fresh vendor teams in Europe.

- Status: In production · Amazon Fresh EU
- The problem: My own team was jumping between 10+ tools to write reports and catch issues with their brands.
- What I did: I started building the fix from scratch with one teammate. Today I lead the product and coordinate a 10-person tech team.
- What happened: An AI operating model that changed how 80 people work and aligned ways of working across 5 countries.
- people working differently: 80 (November 2026)
- countries, same ways of working: 5 (November 2026)
- agentic skills: 10+ (November 2026)
- automated tests: 1000+ (November 2026)
- Data: SQL → semantic model, with data contracts, validated against the source of truth
- AI: 10+ skills (MCP) that draft business reviews, deep dives and vendor reports
- Alert: Ranked by financial impact, urgency and persistence, with a proposed plan
- Approval: A person reviews and approves. Only then does anything run
- Audit: Every step in a tamper-evident log

### Planeo — Your social life, in one app.

Find real plans near you, join with one swipe and meet the people going. Clubs and organizers fill spots and get paid without building anything.

- Status: On the App Store and Google Play
- The problem: A city's plans get lost between stories and WhatsApp groups, and whoever runs them ends up reconciling lists and payments by hand.
- What I did: I designed and built it alone, with Claude Code as the main engineer: both apps, the database, payments, the organizer panel and the website.
- What happened: Live on both stores and used by real clubs. Garbán Woman Lab filled all 4 of its plans, 62 of 62 spots, with 297 waitlist sign-ups.
- Discover: What's on in your city this week, with spots and price up front.
- Join: Swipe and you're in. With a QR to get in and what you might like next.
- Search: By category, by city, or verified organizers only.
- The feed: Your people's photos after every plan.
- Stack: React Native · Expo, Supabase (Postgres, RLS, Edge Functions), Stripe Connect, Apple Wallet, Claude Code
- users: 1427 (November 2026)
- plan sign-ups: 938 (November 2026)
- out of 5, from 136 reviews: 4.85 (November 2026)
- organizers: 18 (November 2026)
- App Store: https://apps.apple.com/app/planeo/id6760037863
- Google Play: https://play.google.com/store/apps/details?id=com.planeo.app

### Ruki — A coach that knows you.

It reads what you run on your watch, knows how much load you can take and writes a plan that grows with you. Ask it anything and it answers like your coach would.

- Status: Private beta
- The problem: A running plan is a template: it doesn't know you played tennis on Tuesday or that your shin hurts.
- What I did: A tool-using agent on Claude with RAG over a curated coaching corpus. It reads each runner's data and writes their plan from their real training load (ACWR).
- How I know it works: Evals as invariants: 7 SQL audits, 100 synthetic athletes and 50 runners simulated over a year. If they fail, the engine does not ship.
- It talks: Tell it your shin is sore and it swaps the hard session for an easy run, without breaking your week.
- It reads you: Strava, Polar and Apple Health come in on their own. Tennis counts as load too.
- It plans: Join your club run and the plan rearranges itself around it.
- Stack: Claude API, RAG on pgvector, Edge Functions, Strava · Polar · Apple Health, React Native · Expo
- activities read from watches: 2200+ (November 2026)
- messages with the coach: 740+ (November 2026)
- audits guarding the engine: 7 (November 2026)
- synthetic athletes: 100 (November 2026)

## KM 4 · How I build

I built Planeo and Ruki with Claude Code as the main engineer. What counts isn't the prompt, it's everything around it.

- Specs in the repo: Every project has its CLAUDE.md with rules, architecture and limits: what an agent must know before touching anything.
- Hooks and permissions: Policies for what it can do without asking and what it cannot. Nothing reaches production without confirmation.
- Subagents and reviews: A written review loop per change. Whoever reviews isn't whoever wrote it.
- Custom skills: Beta diagnostics, incident response and product metrics: they run the same way every time.
- Agents: Tool use, RAG, evals and guardrails, human-in-the-loop, Claude API, Claude Code, MCP
- Engineering: TypeScript, Python, SQL, React, Electron, React Native, Supabase, Stripe
- Data & product: Semantic models, QuickSight, R, pandas, discovery, specs, roadmap, launch
- Languages: Spanish (native), English (fluent)

## Career

- Jul 2026 — now · AI Product Manager · Amazon Fresh EU · Madrid. I lead AVS 2.0 and coordinate a 10-person tech team. A shared AI operating model for 80 people in 5 countries. 10+ agentic skills, validated against the source of truth. Ranked alerts, and actions only with human approval and an audit log.
- Oct 2025 — Jun 2026 · Brand Specialist · Amazon · Madrid. Ran large brands end to end: logistics from the vendor to the customer, catalogue and promotions. This is where I started building AVS 2.0, to fix my team's tooling.
- Mar — Oct 2025 · Retail Vendor Manager Intern · Amazon · Madrid. Onboarded vendors, negotiated their terms and set up their catalogue and logistics inside Amazon.
- Aug — Dec 2024 · Exchange semester · South Korea. A semester abroad, in my final year of university.
- Jan — Jul 2024 · Safety Research & Technical Affairs · Toyota Motor Europe · Brussels. In safety R&D, I led a project on the diversity of automotive study data, so that what gets built on it protects as much of society as possible without leaving minorities behind. Product owner of its data tool. Briefed product teams on the EU AI Act.
- Jun — Aug 2022 · Electrification · Kia Spain · Madrid. The Wible car-sharing fleet and EV charging projects with Iberdrola. My first professional role.
- 2020 — 2025 · Double degree, Business Analytics & Business Administration · Universidad Francisco de Vitoria · Madrid. Analytics honors program. Integral Leadership Program with Notre Dame. Exchange at Emory, Goizueta (Atlanta, 2022).

## Links

- GitHub: https://github.com/manueljimenez-creator
- LinkedIn: https://www.linkedin.com/in/manuel-jimenez-carmona/
- Email: manuel.jimenez@planeoeo.com