Roger Guerrero Plana
~/ai-consultant $ llm · rag · production_

I take generative AI from demo to production, with auditing and security built in.

I co-founded a medical SaaS and built its entire platform, from the LLM and RAG reasoning engine to the backend, payments and deployment. I have automated a law firm with human review and built the guardrails of an emotional support AI. Real production where getting it wrong is expensive.

Let's talk about your project See case studies
[email protected] · guided demo
Medical SaaSclinical AI platform in production, built end to end as co-founder and CTO
Law firm~70% of manual work saved with AI the lawyer reviews and validates
Therapeutic VRemotional support AI engine with safety guardrails, in Mexico and the US
GDPRsensitive health, legal and wellbeing data handled with compliance by design
// case studies

AI in production where getting it wrong is expensive

Healthcare, legal and emotional wellbeing: delicate domains, sensitive data and decisions with consequences. That is where I have built.

TMDream, clinical decision support

Thervia HealthTech · Co-founder & CTO · 2024 to date

LLM + RAGOwn GPUGDPR

It helps healthcare professionals assess orofacial pain and temporomandibular disorders. I designed and built the entire platform, and I still operate it in production.

36institutions using the platform
84real registered users
~300clinical cases in the regression suite
100%self-hosted, no third-party APIs
  • Clinical reasoning engine with LLM and RAG, using an open model self-hosted on our own GPU: clinical data never leaves controlled infrastructure.
  • Conversational clinical interview in 3 languages, reports for clinician and patient, telemedicine with native scheduling, Stripe payments and Google Meet video visits.
  • Clinical quality guarded by a regression suite and periodic adversarial audits of the reasoning (false positives and negatives, red flags).
Python · FastAPINext.jsDockerStripeSEO 18 countriesContinuous auditing
~70%

less manual work at a law firm

SyG Abogados (Barcelona) · Civil law · 2025-2026

An AI that processes the firm's workload overnight and delivers ready drafts of civil claims in the morning. The lawyer only reviews and validates.

  • Human in the loop by design: no document goes out without human validation.
  • Private, internal firm data handled under strict confidentiality.
  • With legal judgement of my own: I hold a Law degree from the UB.
Human in the loopPrivate dataDocument automation
VR

Emotional support AI with guardrails

Therapeutic immersion program · Mexico & US

Conversational engine for emotional support embedded in a personal transformation program combining virtual reality and psychological accompaniment.

  • Built the conversational AI engine and its guardrails.
  • Clear scope limits and careful handling of sensitive situations.
  • A domain where restraint matters more than fluency.
GuardrailsConversational AITherapeutic VR

Pieces built inside these projects:

Corporate website with international SEO Booking & telehealth system Stripe payments Google Meet video visits Google sign-in (OAuth) Admin dashboards Automated PDF reports Multilingual RAG Overnight batch processing

Fewer demos. More production.

// services

How I can help

I work freelance, per project or embedded in your team as a hands-on technical profile. Everything I offer, I have already done in a real product running in production.

/01

LLMs & RAG to production

From use case to working system, not to a pretty demo.

  • RAG pipelines with hybrid, multilingual search, and agents
  • Model and architecture choice to match your case and budget
  • Integration with your product via API, with quality metrics
/02

Self-hosted AI & private APIs

Your AI on your infrastructure: maximum privacy, cost under control.

  • Deployment of open models on your own server or dedicated GPU
  • Private inference APIs for your applications
  • Data never leaves your infrastructure, key when it is sensitive
/03

AI system audits & evaluation

Does your AI actually answer well? I measure it before your users find out.

  • Automated evaluation and regression suites
  • Adversarial audits: false positives, hallucinations, bias
  • Reasoning traceability in critical domains

Small engagements to start · fixed price

A

An assistant that answers on your documents

Manuals, protocols, catalogues or internal policy. Your team asks in plain language and gets the answer with its source cited.

What backs it: TMDream's multilingual RAG engine, in production.
B

Document extraction into a spreadsheet

Invoices, contracts or forms someone types by hand today, turned into structured data, with human review before anything counts as final.

What backs it: the law firm's document automation, human in the loop.
C

Meeting notes and summaries, on your server

Transcription, summary and action list for every meeting. The audio never leaves your infrastructure, which no cloud tool can promise you.

What backs it: self-hosted models on our own GPU since 2024.
D

Express audit of your AI

Already have an assistant and don't know if it answers well? Case suite, adversarial testing, a report of the failures and a prioritised fix plan.

What backs it: periodic audits of TMDream's reasoning and its suite of nearly 300 cases.
E

AI training: companies, universities and schools

Hands-on sessions tailored to the audience: company teams, university faculty and students, or secondary schools. What AI does well, what it does badly, how to use it with judgement and where the limits are.

What backs it: master's lecturer at the University of Barcelona and educational innovation coordinator for 300+ students.

Also, when the project calls for it:

Backend & full product

APIs with Python and FastAPI, frontend with Next.js, payments, scheduling, Docker and deployment.

Websites, CRMs & custom apps

Corporate websites with technical SEO, internal dashboards, hosting and maintenance.

Security & compliance

Application and server audits, hardening, access control and GDPR.

AI training & adoption

Hands-on training for teams, faculty and students, plus use-case strategy. Teaching at the University of Barcelona.

// how i work

Three steps, no hype

I work per fixed project or monthly dedication, always with reviewable deliveries.

01

Honest assessment

I look at your case and tell you what I would do, how long it takes and the risks. If you don't need AI to solve it, I'll say that too.

02

Build in deliveries

Working pieces from week one, not promises. You see and validate every step on your own case.

03

Production and guarantees

Deployment, a test suite that protects every change, security and documentation so your team can maintain it.

// education

Credentials

Working on an AI project?

Tell me what you want to build, or what isn't working. I'll reply with an honest assessment: what I would do, how long it would take, and whether you actually need AI to solve it.

Email me[email protected] LinkedInroger-guerrero-plana