ZTABS fine-tunes large language models for Barcelona enterprises -- domain-specialized models for pharmaceutical clinical documentation, multilingual hospitality operations, Catalan-Spanish-English enterprise NLP, and Southern European market intelligence across the Mediterranean business hub.
Barcelona's LLM fine-tuning requirements reflect the city's unique position at the intersection of pharmaceutical innovation, Mediterranean tourism, and one of Europe's most linguistically complex business environments. The BioRegion of Catalonia -- over 1,100 life sciences companies -- generates massive volumes of clinical and regulatory documentation that combines scientific terminology, regulatory language, and multilingual content in ways that general-purpose LLMs handle poorly. A clinical trial protocol written in Barcelona references AEMPS (Spanish regulatory) requirements, EMA (European) guidelines, and ICH (international) standards, using medical terminology in Spanish and English with regulatory references that span multiple jurisdictions. Foundation models that produce fluent medical text frequently misapply regulatory requirements from one jurisdiction to another or use American clinical terminology where European conventions differ. The linguistic dimension is particularly compelling in Barcelona. The city operates in three primary languages -- Catalan, Spanish, and English -- with French, German, Italian, and other European languages common in business interactions. Catalan is not a dialect of Spanish; it is a distinct Romance language with its own grammar, vocabulary, and literary tradition, spoken by approximately 10 million people. Fine-tuning for Barcelona enterprises must handle Catalan as a first-class language rather than treating it as Spanish variant -- an error that general-purpose models frequently make, producing output that mixes Catalan and Spanish grammatical structures or applies Spanish vocabulary where Catalan terms differ. For the Generalitat de Catalunya (regional government), universities, healthcare institutions, and many private companies, Catalan-language capability is not optional but essential for operations and regulatory compliance. Pharmaceutical documentation presents the most technically demanding fine-tuning use case. Clinical trial documents follow ICH-GCP guidelines with jurisdiction-specific adaptations: Spanish-language informed consent forms follow AEMPS requirements that differ from FDA or MHRA specifications, adverse event reporting uses MedDRA coding with language-specific verbatim descriptions, and study reports must satisfy both AEMPS and EMA expectations. A general-purpose LLM processing a Spanish-language adverse event report may apply incorrect MedDRA coding conventions, misclassify event severity using criteria from the wrong regulatory framework, or generate follow-up queries using American clinical trial terminology that Spanish site coordinators would not recognize. Hospitality documentation adds a multilingual dimension where fine-tuning delivers immediate operational value. Barcelona's 2,800+ hotel properties interact with guests in 8-12 languages daily, generating guest communications, reviews, feedback forms, and internal operations reports. Fine-tuning models on hospitality-specific corpora across multiple languages produces capabilities that generic translation tools cannot match -- understanding that a guest complaint about "la climatizacion" requires a maintenance response rather than a front desk response, that "el parking" is commonly used in Barcelona hotel Spanish despite being a Spanglish borrowing, and that cultural expectations around service recovery differ between German, Japanese, and American guests. AEPD compliance and the emerging EU AI Act framework create specific requirements for fine-tuning pipelines. Barcelona enterprises must ensure that training data handling complies with Spanish data protection law (LOPD-GDD) and GDPR, and that fine-tuned models used in automated decision-making maintain the transparency and explainability requirements that AEPD enforces.
Barcelona's LLM fine-tuning demand concentrates across four sectors where multilingual domain specialization creates measurable value. Pharmaceutical and clinical research is the anchor vertical. The BioRegion of Catalonia's 1,100+ life sciences companies generate documentation spanning drug discovery, clinical trials, regulatory affairs, pharmacovigilance, and commercial operations. Fine-tuning use cases include clinical document processing (training models on Spanish and English clinical trial protocols, informed consent forms, case report forms, and study reports following AEMPS and EMA conventions), pharmacovigilance automation (fine-tuning on adverse event reports in Spanish to improve MedDRA coding accuracy and case processing efficiency), regulatory submission preparation (training models on AEMPS and EMA submission formats to automate document compilation and quality checking), and medical literature analysis (fine-tuning on Spanish and English medical publications to extract clinical findings relevant to specific therapeutic areas). CROs (contract research organizations) operating from Barcelona add demand for fine-tuning on multi-country clinical trial documentation across Southern European regulatory frameworks. Tourism and hospitality constitutes the second major vertical. Barcelona's hospitality sector processes guest interactions in multiple languages with domain-specific terminology. Fine-tuning addresses guest communication (training models on hospitality-specific language across Catalan, Spanish, English, French, German, Italian, Japanese, and Korean -- understanding that hotel-specific vocabulary and cultural conventions differ from general language), review analysis (fine-tuning on hotel review corpora to extract actionable operational insights from multilingual guest feedback), and operational documentation (training models on housekeeping, maintenance, F&B, and front office documentation that uses industry-specific Spanish terminology). Legal services represent the third pillar. Barcelona's legal market produces documents in Spanish, Catalan, and English following Spanish civil law traditions. Fine-tuning addresses contract analysis (training on Spanish commercial contracts, condiciones generales de contratacion, and the specific provisions of the Codigo Civil and Codigo de Comercio), corporate documentation (fine-tuning on Spanish corporate governance documents, actas, and the specific formats required by the Registro Mercantil), and regulatory compliance (training models on AEPD filings, CNMV securities documentation, and industry-specific regulatory requirements). Technology and startups round out the top four. Barcelona's tech ecosystem needs fine-tuned models for Catalan and Spanish product documentation, customer communications, and content generation at scale.
Our Barcelona LLM fine-tuning methodology addresses the multilingual complexity and domain specialization that characterize Mediterranean enterprise. We begin with a Multilingual Domain Assessment: mapping the specific languages, registers, and domain terminology the client uses, identifying where general-purpose models fail on each language, and designing training strategies that build genuine multilingual capability rather than treating each language independently. For pharma clients, this means understanding that clinical terminology has language-specific conventions -- "evento adverso grave" in Spanish follows AEMPS reporting conventions that differ from FDA "serious adverse event" definitions, and the model must apply the correct regulatory framework based on document context. Training data curation for Barcelona clients prioritizes the multilingual domain content that foundation models handle poorly. We assemble training corpora that include Catalan, Spanish, and English domain text in natural proportions, preserving the code-switching patterns that occur in Barcelona professional environments (meetings that shift between Catalan and Spanish, documents that use English technical terms within Spanish text). For pharma clients, we curate clinical documentation from Spanish and Southern European regulatory contexts, ensuring the model learns AEMPS-specific requirements rather than defaulting to FDA or MHRA conventions. Model architecture selection evaluates multilingual capability as a primary criterion. Barcelona deployments typically require strong performance across Catalan, Spanish, and English at minimum, with additional European languages based on the client's market. We select foundation models with demonstrated Romance language capability and fine-tune in ways that improve domain accuracy across all target languages simultaneously. For Catalan specifically, we supplement training with Catalan-language domain corpora to prevent the model from treating Catalan as a Spanish dialect. Evaluation uses native-speaker domain experts for each target language. Catalan output is evaluated by native Catalan speakers with domain expertise (not Spanish speakers who "understand Catalan"), Spanish output by native Spanish speakers familiar with the specific professional domain, and English output by domain experts who can assess both language quality and domain accuracy. This rigorous multilingual evaluation ensures that fine-tuned models meet professional standards in every language they produce.
Pharmaceutical / Clinical Research
A Barcelona-headquartered pharmaceutical company managing 18 active clinical trials across Spain, Italy, Portugal, France, and Greece needed to improve the efficiency of adverse event processing in their pharmacovigilance department. The company received approximately 3,200 Individual Case Safety Reports (ICSRs) annually from investigator sites, each requiring: verbatim description extraction from the source language (Spanish, Italian, Portuguese, French, or Greek), MedDRA coding of the adverse event using the Lowest Level Term (LLT) and mapping to Preferred Term (PT), System Organ Class (SOC), and High Level Group Term (HLGT), seriousness assessment according to ICH-E2A criteria, causality assessment preparation (assembling the clinical information required for the medical reviewer's causality determination), and regulatory reporting timeline assessment (determining whether the case meets expedited reporting criteria under AEMPS, AIFA, Infarmed, ANSM, or EOF requirements). The pharmacovigilance team of 8 safety scientists processed ICSRs with an average turnaround of 6.4 business days from receipt to quality-checked case entry. MedDRA coding accuracy averaged 84%, with errors concentrated in cases where verbatim descriptions used colloquial medical language or where the same symptom was described differently across languages. Regulatory reporting timeline compliance was 94% -- below the company's 99% target -- with late reports primarily caused by processing delays during volume spikes (typically during interim data reviews when sites submitted batched adverse event reports).
We developed a fine-tuned LLM for multilingual pharmacovigilance processing: (1) Training Corpus -- assembled from the company's safety database: 18,000 historical ICSRs in Spanish (42%), Italian (22%), Portuguese (14%), French (16%), and Greek (6%). Each ICSR included the original verbatim adverse event description, the MedDRA coding assigned by safety scientists, seriousness assessment, and regulatory reporting determination. Senior safety scientists annotated 2,400 cases with detailed coding rationale, explaining why specific MedDRA terms were selected and how verbatim descriptions in each language mapped to standardized terminology. (2) MedDRA Coding Model -- fine-tuned specifically on the mapping between multilingual verbatim descriptions and MedDRA terms. The model learned language-specific medical colloquialisms: that "mareo" in Spanish could map to PT "Dizziness," "Vertigo," or "Nausea" depending on the clinical context described, that Italian "malessere generale" maps to PT "Malaise" rather than the literal "general unwellness," and that Portuguese medical terminology uses Latin-derived terms that sometimes differ from Spanish equivalents despite the languages' similarity. The model outputs a primary MedDRA coding with confidence score plus alternative codings when ambiguity exists, enabling safety scientists to make informed coding decisions. (3) Regulatory Assessment Module -- trained on AEMPS, AIFA, Infarmed, ANSM, and EOF expedited reporting criteria. The model assesses each case against the applicable regulatory requirements based on the reporting country and determines whether expedited reporting (7-day or 15-day) is triggered. This assessment is presented to the safety scientist for confirmation rather than executed autonomously, maintaining the human oversight required by pharmacovigilance regulations. (4) Integration -- deployed within the company's safety database, processing incoming ICSRs as they are received from sites and presenting pre-coded cases to safety scientists for review. The system flags high-priority cases (potential SUSARs) immediately, regardless of queue position.
MedDRA coding accuracy improved from 84% to 94.2% on the primary coding suggestion, with the correct MedDRA PT appearing in the top-3 suggestions in 98.6% of cases. Accuracy improvements were consistent across all five languages. ICSR processing time decreased from 6.4 business days to 2.1 business days, with the model performing initial coding and assessment within minutes of case receipt and safety scientists reviewing pre-coded cases rather than coding from scratch. Regulatory reporting timeline compliance improved from 94% to 99.4%, exceeding the company's 99% target. The improvement was driven by immediate identification and prioritization of potentially expeditable cases, eliminating the processing delays that had caused late reports. Volume spike handling improved dramatically: during a major interim data review that generated 340 ICSRs in 10 days (versus the normal rate of ~60 per month), the model processed all incoming cases within hours, enabling the safety team to maintain timely processing without emergency overtime. The company estimated annual value at EUR 1.4 million: EUR 620,000 in pharmacovigilance team efficiency, EUR 380,000 in regulatory compliance improvement (reduced risk of regulatory action from late reporting), and EUR 400,000 in data quality improvement (more accurate MedDRA coding enabling better safety signal detection). The safety director noted that the fine-tuned model's understanding of Southern European medical language was "transformative -- it understands how Spanish doctors describe symptoms in ways that no previous tool has matched."
We clean, deduplicate, and structure your training data into high-quality instruction-response pairs. Quality data is the single biggest factor in fine-tuning success.
Fine-tune GPT-4o Mini and GPT-3.5 Turbo through OpenAI's API with systematic hyperparameter optimization, validation splits, and automated evaluation.
Fine-tune Llama 3, Mistral, Phi, and other open-source models using LoRA, QLoRA, and full fine-tuning on cloud GPUs or your own infrastructure.
Rigorous evaluation against your specific tasks with automated benchmarks, human evaluation, and A/B testing against base models to quantify improvement.
Align model outputs with human preferences using DPO (Direct Preference Optimization) and RLHF techniques for better quality and safety.
Deploy fine-tuned models via OpenAI, vLLM, TGI, or Ollama with optimized inference, batching, and auto-scaling for production workloads.
Barcelona's LLM fine-tuning market includes pharma technology companies (IQVIA, Oracle Health Sciences), Spanish AI companies, and local technology consultancies. Pharma technology companies offer pre-built NLP tools for pharmacovigilance and clinical documentation but typically lack fine-tuning capability for Southern European languages and regulatory frameworks. Spanish AI companies provide general language processing but may lack pharmaceutical domain expertise. Our differentiation is fine-tuning at the intersection of multilingual capability (Catalan, Spanish, English, and Southern European languages), pharmaceutical domain expertise, and the regulatory knowledge required for AEMPS and EMA compliance.
LLM fine-tuning in Barcelona typically ranges from EUR 40,000 to EUR 200,000 depending on domain complexity, multilingual requirements, and deployment architecture. Domain assessment and corpus strategy run EUR 12,000-25,000. Single-task fine-tuning ranges from EUR 40,000-90,000. Multi-task multilingual fine-tuning with production deployment ranges from EUR 90,000-160,000. Enterprise programs with continuous learning run EUR 160,000-200,000. Ongoing model maintenance runs EUR 3,000-10,000 monthly. Barcelona pricing is competitive within Europe -- typically 25-35% below London or Munich for equivalent technical quality. We offer a 2-week proof-of-concept at EUR 10,000-20,000 demonstrating accuracy improvement on multilingual domain documents.
Verified reviews from real client engagements — sourced from our public testimonial archive and Clutch profile.
My experience is throughout positive. Communication, service, the short response times and the flawless execution of a challenging topic was absolutely great. ZTABS is definitely my first choice again.
Christian Neff
Bank Software Advisory · Bank Software Advisory
Fintech
Fantastic Agency! I couldn't fault them even if I tried. They always go above and beyond to meet your expectations and always produces quality work. Thank you ZTABS.
Stephanie Kal
CEO · Beauty Finder Australia
Marketplace
It has been great working with ZTABS. They bounce off the ideas along the way. Amazing Experience.
Joel Rowe
CEO · Drill Quoter
Marketplace
We don't just contract. We ship and operate our own software. 23 products in production.
Common questions about LLM fine-tuning for Barcelona businesses
We treat Catalan as a first-class language, not a Spanish dialect. Fine-tuning includes Catalan-language domain corpora, evaluation by native Catalan-speaking domain experts, and training strategies that prevent the model from mixing Catalan and Spanish grammatical structures. For organizations that operate in Catalan (Generalitat, universities, healthcare institutions), this produces models that generate professionally appropriate Catalan rather than Catalan-flavored Spanish.
ZTABS is a remote-first LLM fine-tuning agency serving Houston businesses — including data pipeline & curation, openai fine-tuning, open-source model training. We work with Energy & Oil/Gas, Healthcare & Biotech, Aerospace & Defense companies in Houston, TX via timezone-aligned engineers and async workflows; we do not have a local office, and we are explicit about that with every client.
LLM Fine-Tuning in New York, NYZTABS is a remote-first LLM fine-tuning agency serving New York businesses — including data pipeline & curation, openai fine-tuning, open-source model training. We work with Finance & Fintech, Media & Advertising, Fashion & Retail companies in New York, NY via timezone-aligned engineers and async workflows; we do not have a local office, and we are explicit about that with every client.
LLM Fine-Tuning in Los Angeles, CAZTABS is a remote-first LLM fine-tuning agency serving Los Angeles businesses — including data pipeline & curation, openai fine-tuning, open-source model training. We work with Entertainment & Media, E-commerce & DTC Brands, Gaming & AR/VR companies in Los Angeles, CA via timezone-aligned engineers and async workflows; we do not have a local office, and we are explicit about that with every client.
Web Development in Barcelona, SpainZTABS is a remote-first web development agency serving Barcelona businesses — including full-stack development, progressive web apps, api development. We work with Mobile & Telecom, Travel Tech, HealthTech companies in Barcelona, Spain via timezone-aligned engineers and async workflows; we do not have a local office, and we are explicit about that with every client.
Web Design in Barcelona, SpainZTABS is a remote-first web design agency serving Barcelona businesses — including ui/ux design, responsive design, custom interfaces. We work with Mobile & Telecom, Travel Tech, HealthTech companies in Barcelona, Spain via timezone-aligned engineers and async workflows; we do not have a local office, and we are explicit about that with every client.
AI Development in Barcelona, SpainZTABS is a remote-first AI development agency serving Barcelona businesses — including llm integration & fine-tuning, ai agents & automation, rag & knowledge systems. We work with Mobile & Telecom, Travel Tech, HealthTech companies in Barcelona, Spain via timezone-aligned engineers and async workflows; we do not have a local office, and we are explicit about that with every client.
LLM Fine-Tuning ServicesLearn more about our LLM fine-tuning services nationwide.
Partner with ZTABS for expert LLM fine-tuning in Barcelona. Get a free consultation today.
