Applied AI Engineer & Technical Product Leader.

About Manuel Mezo

Aerospace and Mechanical Engineer (UC3M · TU Delft) turned applied AI builder, with an 8+ year track record of solving multi-million-dollar operational challenges at Amazon and McKinsey. Now independently shipping production AI systems: end-to-end RAG pipelines, autonomous LLM agents, computer vision, and multi-agent simulations. I combine deep operational instinct for what AI needs to do for real users with hands-on engineering capability across the full stack: Python, FastAPI, cloud infrastructure, ML fine-tuning, and real-time deployment. Seeking an AI Engineer or Robotics Engineer role at a startup, or a Technical PM position where engineering depth is an asset. Fluent in Spanish (native), English (bilingual), and French (B2).

Professional Experience 🧑🏻‍💻

📄 Download CV in PDF

Feb 2026 – Present

Manager, EU Transportation Forecasting
Amazon (Madrid, Spain)

• Leading a team of 6 BIEs and Program Managers to deliver a strategic forecasting roadmap targeting >$10M/year in cost savings via ML bias correction, simulation modules, and automated health checks.
• Co-own the technical roadmap with the Forecasting Tech team, driving a 150 bps WAPE reduction across all EU network segments.
• Own end-to-end production of weekly demand forecasts that optimize transportation and labor capacity planning across all EU logistics.

Sept 2023 – Feb 2026

Sr Program Manager, Supply Chain Planning
Amazon (Madrid, Spain)

• Invented and implemented a 52-week long-range forecasting methodology for EU truckload demand, directly informing $140M in CapEx decisions for 2025.
• Extended the inbound forecasting horizon to 12 weeks, improving planning coverage for 800k annual labor hours.
• Improved inventory transfer truck fill-rate modeling (inaccuracy cut from 35% to 17%), saving ~$1.8M/year by eliminating unnecessary truck additions and cancellations.
• Identified and resolved a critical logic defect in the automated scheduling system using SQL-based analysis and Python scripting, protecting millions in avoidable operational costs.

Aug 2022 – Jul 2023

Operations Consultant
McKinsey & Company (Madrid, Spain)

• Redesigned and automated production capacity planning for a global pharmaceutical manufacturer, rolled out across 6 plants and 13 production lines.
• Optimized production batch sizes across 9 plants, saving $800k/year (1.1% of conversion cost).
• Recovered 2.5 months of delivery delays for a train manufacturer, eliminating $1M in contract penalties.

Sept 2017 – Aug 2022

Sr PM, Ops Excellence & Quality
Amazon (Madrid / Luxembourg)

• Led planning and execution of multi-item order shipping capabilities across 10 3PL sites, generating $10M+ in transport savings.
• Designed network-wide inventory audit programs across 25+ warehouses, improving Inventory Record Defect Rate by 50%.
• Recovered a severe inventory crisis from 120k to <10k DPMO in 8 weeks through process redesign.
• Led root cause investigation and settlement negotiation on a $7M disputed inventory loss.

Tech Stack 🛠️

Education 🎓

2017

Master's Thesis Exchange
Delft University of Technology

• 6-month exchange to elaborate Master's thesis: "Open- and closed-loop Doppler data modelling for spacecraft tracking".
• Part-time volunteering engineer at Formula Student Team Delft.

2015 – 2017

MSc in Aerospace Engineering
Universidad Carlos III de Madrid

• Focus on integral aircraft and satellite design, aeronautical and space propulsion, aerospace navigation and control.
• Exchange program at TU Delft (Netherlands).

2011 – 2016

BSc in Aerospace Engineering
BSc in Mechanical Engineering
Universidad de León

• Basketball player and referee.

📄 Download CV in PDF

GenAI Projects 🚀

❮

SaaS Support Agent (InteractiveAI)

End-to-end customer support agent built on the InteractiveAI platform covering the full lifecycle: Build context, Connect tools & secrets, Deploy, Govern traces, and Improve with datasets.

View full project →

💬 Portfolio Chat Agent

End-to-end RAG system built with LangChain, ChromaDB, and Gemini 2.5 Flash. Ask about my background, experience, and projects. The agent answers grounded in my actual documents.

View full project →

🧾 3-Way Invoice Matching Agent

Autonomous Accounts Payable (AP) back-office agent built with Gemini 2.5 Flash, Next.js, and Neon Postgres. Reads messy vendor invoices, cross-references Purchase Orders and WMS receipts, then approves, flags, or escalates. Includes a full live reasoning trace.

View full project →

📞 Omnichannel AI Customer Service Platform

Full-stack omnichannel CS simulation integrating AI voice calls (Retell.ai), live web chat (Voiceflow), and automated email (Resend). Real-time order lookups and unified conversation handling across all three channels.

Try it out!

🩺 X-Ray Medical Agent

Autonomous diagnostic agent powered by Computer Vision for real-time medical image analysis.

Try it out!

🏛️ MonumentScout

Location-based discovery app using Geolocation APIs and AI-powered explanations to explore nearby historical monuments.

Try it out!

📦 WMS Prototype

Mobile-first inventory system leveraging on-device Camera Vision for barcode scanning and real-time cloud synchronization.

Try it out!

🎙️ Voice Booking Agent

AI voice assistant using Web Speech API and LLM orchestration to book and manage your Google Calendar appointments via natural language voice conversation.

Try it out!

💻 My Github profile

Public repository of my ongoing experiments in AI prototyping, web/mobile applications, and AI-assisted software development.

Check it out!

🌐Web CV

AI-developed interactive portfolio showcasing my professional accomplishments, education, and AI side hustles.

Thanks for watching!

❯

Computer Vision projects 👁️

❮

🔌 YOLOv8 Inference API: Production Deployment

Productionized YOLOv8 as a cloud REST API: async task queue for multi-user concurrency, daemon-thread memory management against container OOM crashes, Docker containerization, and live on Hugging Face Spaces.

View End-to-End Case Study →

🎯 Part I: Custom YOLOv8 Object Detection (Fine-Tuned)

Custom object detection pipeline: annotated dataset in Roboflow, augmentation for motion blur & lighting variance, YOLOv8n fine-tuning over 100 epochs (~97% mAP@50), and real-time multi-object tracking on a conveyor belt.

View full project →

📦 Part II: Multi-Object Tracking & Spatial Logic (ByteTrack + OpenCV)

Advanced spatial logic built on top of a real-time object tracking engine. Implements virtual polygon pick-zones (point-in-polygon logic) and vertical throughput tripwires to count and register custom items on a moving conveyor belt.

View full project →

🚦 Realtime Traffic data logger and analyzer

This project consists of a browser dashboard displaying a live video feed in which virtual gates can be defined to track, classify, and count vehicles and pedestrians in real-time, featuring live telemetry graphs and CSV export.

👁️ VisionAI

Real-time video analysis powered by Gemini 2.5 Flash. Detects and tracks objects via natural language queries with live bounding boxes.

Try it out!

❯

Robotics side ventures 🤖

❮

🦾 3D Printed Robotic Arm

Physical 4-axis robotic arm integrated with a cloud-based Vision-Language-Action (VLA) reasoning engine. Translates natural language voice commands via Gemini 1.5 Pro into precise local mechanical movements.

View full project →

🏎️ ROS2 Wall Follower

Autonomous navigation logic using Python and multithreading, parsing LIDAR data to track and successfully follow walls in a simulated environment.

Check it out!

🤖 Self-Balancing Robot (Real)

Physical deployment of my simulation built on Arduino hardware. Implements PID control, IMU filtering, and motor actuation in the real world.

Check it out!

⚖️ Self-Balancing Robot (Sim)

ROS2 simulation of a self-balancing robot built in Gazebo. Focuses on sensor modeling, odometry, and control loop tuning.

Check it out!

❯

Personal Interests 🤓

Click an interest below to see a related photo!