About

I move between disciplines to find the structure underneath them, above all the way a system changes when reality pushes back.

Once the constraints are set, the rest is just interpolation.

I move between disciplines to find the structure underneath them, above all the way a system changes when reality pushes back.

The point of view. I trained in industrial engineering, where a design eventually meets its materials and the ways it can fail. That perspective stayed with me as my interests expanded into software and AI. I am less interested in collecting domains than in understanding what transfers between them. A bottleneck in a supply chain and a cognitive constraint are different problems, but they reward the same habit of looking for structure. Most things here exist because they became too interesting to leave alone. Apart from all this, a more biological curiosity: how the brain learns, which I follow in Lobeworks/17 (https://infraphysics.net/lobeworks-17).

Field of view: ML, networks, tools, economics, web dev, infrastructure, mathematics, computer science.

Working principles. Look for the constraint before optimizing the surface. Make complexity traversable; do not pretend it is simple. Use models to think, then let reality correct the model. Build enough to discover what explanation alone cannot. Keep what survives curiosity; discard the rest.

If you are working on a difficult system or an idea that crosses boundaries, I would probably like to hear about it.

Experience and CV (in full at https://infraphysics.net/about/cv):

Yago Mendoza | Industrial and applied AI engineer

I am an industrial engineer and systems-oriented generalist with particular depth in the evaluation and deployment of applied AI, and in end-to-end product engineering.

I use that breadth to give ambiguous problems a structure and build systems that work in practice.

Selected evidence:

70 → 93%: DxGPT evaluated diagnostic accuracy, improved through model selection, inference optimization and safety guardrails.

≈700K: Queries per year served by the clinical AI system supported by that evaluation capability.

0 → shipped: TrialGPT built end to end: document parsing, clinical extraction, matching, evaluation, UX, backend and Azure deployment.

Core capabilities:

- AI evaluation & model selection

- Production deployment & reliability

- End-to-end product engineering

- Systems integration under operational constraints

Experience:

Foundation 29 | Applied AI Engineer · Clinical AI | 2025 – Present

Building and operating clinical AI systems where model behavior requires measurable evidence, resilient infrastructure and human review.

- Built the organization’s first GenAI evaluation capability, turning clinical cases and normalized diagnostic ground truth into benchmarks for model selection and ship/no-ship decisions.

- Helped raise DxGPT’s evaluated diagnostic accuracy from approximately 70% to 93% (about 700K queries a year) through model selection, inference-pipeline optimization, prompting and safety guardrails.

- Shipped TrialGPT end to end: document parsing, clinical concept extraction, trial matching, human review, UX, backend and Azure deployment.

- Built provider-agnostic infrastructure (OpenAI, Azure OpenAI, Claude, Gemini) with multi-region failover and latency and cost optimization.

Leadrank | Product Engineer · Independent | 2026 – Present

Building an AI product that turns unstructured email history into structured commercial intelligence for real-estate workflows.

- Building an AI system that turns unstructured email history into structured contacts, deals, buyer intent and property matches.

- Designed ingestion and extraction for historical and real-time email, with entity resolution and match generation.

- Validating workflows and product demand with agents in Barcelona and Madrid.

UNE | AI Standards Committee · EU AI Act | Feb 2026 – May 2026

Participated independently in Spain’s mirror committee for European and international AI standardization.

- Independent participant in CTN 71/SC 42, Spain’s mirror committee of CEN/CLC/JTC 21.

- Studied prEN drafts and standardization workflows on AI system evaluation, bias management, dataset quality, risk management and conformity assessment.

Sony Europe | ML & Distributed Systems Engineer | 2024

Worked inside Sony Europe’s Brussels ML R&D lab across blockchain optimization, program synthesis and agent evaluation.

- Designed execution-based agent evaluation using compilation, automated tests and network deployment.

- Took the research system end to end, from DSPy and LangChain experiments to orchestration and infrastructure.

- Operated the Hyperledger Besu network, SDK and monitoring stack; selected optimizations entered Sony’s work.

TE Connectivity | Supply Chain Data & Automation Intern | 2023

Applied software and data engineering to operational supply-chain processes.

- Automated inventory, CRM and bill-of-materials reporting with Python and SQL.

- Built internal pipelines and reporting tools to improve data quality and planning visibility.

CIMNE | Safety Research Engineer · EU LASH FIRE | Sep 2021 – Jul 2022

Worked on safety-critical engineering within LASH FIRE, the EU Horizon 2020 project on ro-ro vessel fire safety.

- Co-designed the software architecture of LASH FIRE.

- Designed a vehicle-sorting algorithm that improved simulated fire-safety outcomes by 7%.

- Evaluated early GPT-3 models for technical work before LLMs became mainstream.

Selected work:

- TrialGPT: from a medical report to a shortlist of eligible trials: A clinical-trial matching tool built at Foundation 29 for patients with rare and hard-to-diagnose conditions. TrialGPT reads a case or an uploaded report and checks it against the inclusion and exclusion criteria of trials in ClinicalTrials.gov and the European CTIS register, one clause at a time, keeping a third verdict for what the report does not say. The write-up follows the 2025 build through its failures, then reads how the live product works a year on. (https://infraphysics.net/lab/projects/trialgpt-clinical-trial-matching)

- Model-based fault detection in a two-tank system: Model-based fault detection and isolation on a two-tank hydraulic plant under closed-loop control. Thirteen component relations are reduced by structural analysis to four analytical redundancy relations, and the pattern their residuals form on Simulink runs names the faulty component. Exact signature matching fails on the pump fault; the piece traces that failure and the noise cost of differentiating level measurements before it states what a physical installation would need. (https://infraphysics.net/blog/bits2bricks/two-tank-fault-detection)

- A CRM for a health-tech startup, built solo in 72 hours: A bespoke CRM for a health-tech startup, designed and shipped solo in 72 hours as an experiment in AI-assisted engineering. Its 23 tables are held together by 8 explicit state machines, and a stale-detection clock flags the relationships going cold. (https://infraphysics.net/lab/projects/health-tech-crm)

- How InfraPhysics is built: a site with no server, organised around its wiki: InfraPhysics is a personal site built as one repository and one build: about 2,800 markdown files, a wiki of more than 1,500 notes among them, become static files served with a thin layer of edge functions. This write-up follows one paragraph from the keyboard to its five kinds of reader, then reconstructs from 1,148 commits the decisions that shaped each layer and what each one cost. It began in March as a diary of the first eighteen days and was rewritten in October. (https://infraphysics.net/lab/projects/building-infraphysics)

Contact: [email protected]