AI Agent Development Company in Lisbon

Lisbon agent engagements are priced by tool count and how much review sits around them: a scoped agent proof of concept at Lisbon rates, a production customer support, document extraction, or sales-assist agent, or a multi-agent enterprise rollout with several tool-using agents, role-based human-in-the-loop review, and integration with Salesforce, ServiceNow, SAP, or core banking. Scope drivers include data audit, a baseline RAG pipeline, a tool-using agent prototype, a hosted demo, MLOps, monitoring, prompt injection defences, and an EU AI Act risk classification document. Lisbon rates sit below London, Paris, and Amsterdam while the engineering bench is comparable, which is why VW, BNP Paribas, Cloudflare, and Revolut all relocated engineering here. We give fixed-fee proposals against milestones.

The EU AI Act entered into force on 1 August 2024 and applies in staged phases through 2025 to 2027. Prohibited practices have applied since February 2025, general-purpose AI model obligations since August 2025, and high-risk AI system obligations apply from August 2026 (with extensions to August 2027 for systems embedded into regulated products). Our discovery phase runs an EU AI Act classification on the agent use case (most customer-facing agents are limited-risk with transparency obligations, while agents involved in employment, credit, education access, biometrics, or critical infrastructure are high-risk). High-risk builds ship with the Annex IV technical documentation, a risk management system, data governance documentation, logging, human oversight design, accuracy and robustness testing, and conformity assessment evidence. Limited-risk agents ship with the disclosure obligation that users are interacting with AI and that AI-generated content is labelled.

Yes. The CNPD (Comissão Nacional de Proteção de Dados) is the Portuguese data protection authority and enforces GDPR plus Portuguese national legislation (Lei n.º 58/2019). Every Lisbon engagement opens with a CNPD-aligned data flow map covering categories of personal data, lawful basis, retention, cross-border transfers, automated decision-making profile, and the Article 35 DPIA where personal data feeds the agent. The CNPD has been active on automated decision-making and large-scale processing, so we treat profiling and credit-scoring agents with extra rigour, building in privacy-by-design controls (data minimisation, on-device PII redaction, role-based dashboard access) and the Article 22 GDPR right to human review. We default to EU residency on AWS eu-west-3, Azure West Europe, or GCP europe-southwest1, document SCCs for any non-EEA processor, and produce the Article 30 record.

Yes. We ship customer support agents, KYC document extraction pipelines, AML transaction-monitoring copilots, and internal back-office agents for Banco de Portugal-supervised banks (Caixa Geral de Depósitos, Millennium BCP, Novobanco, BPI) and for the large near-shore engineering hubs of BNP Paribas IT, Natixis Portugal, and Revolut. The engagement runs inside DORA (in application from 17 January 2025), Banco de Portugal Aviso 3/2020 on outsourcing and IT risk, and the EU AI Act high-risk track for credit decisioning and creditworthiness assessment. We default to EU regions only, apply PII redaction at ingest, run a challenger evaluation harness, and produce the third-party register entries that the Banco de Portugal and the European Central Bank's Single Supervisory Mechanism read directly. Prompt injection defences and a human-in-the-loop gate are not optional.

Talkdesk has been a flagship Lisbon platform for cloud contact centres, and several of our clients run agent layers that plug directly into Talkdesk's APIs and webhooks for live agent assist, after-call summarisation, and quality assurance scoring. OutSystems is the dominant low-code platform in many Portuguese enterprises, so we deliver AI agent components as OutSystems Forge modules where the host application is OutSystems-built. Unbabel's machine translation plus human-in-the-loop architecture is a natural fit for multilingual support agents, and we have integrated their endpoints into agent pipelines that serve customers across Portuguese, Spanish, French, Italian, German, and English. Cloudflare Workers AI sits well as the edge inference layer for low-latency agent tasks, and we deploy there when the use case justifies the runtime model.

We default to AWS eu-west-3 in Paris for clients standardised on AWS, with eu-west-1 Dublin and eu-central-1 Frankfurt as alternates, plus Bedrock for hosted Anthropic Claude and Mistral. Azure West Europe in the Netherlands carries Azure OpenAI for clients standardised on Microsoft, with Sweden Central and France Central as alternates. GCP europe-southwest1 in Madrid is the nearest GCP region geographically for Lisbon clients, with europe-west4 in Eemshaven and europe-west9 in Paris as alternates for Vertex AI. Cohere maintains EU endpoints. Mistral's primary endpoints are in Paris and align well with Lisbon's WET to CET timing. Each deployment ships with a documented residency statement that names regions, replication paths, encryption keys (AWS KMS, Azure Key Vault, Cloud KMS), and the data processing agreement that auditors attach to the Article 30 record.

A typical production agent (customer support copilot, KYC document extraction, internal-knowledge RAG over Confluence and Jira, sales-assist agent over Salesforce and HubSpot) takes twelve to twenty weeks from kickoff to production, assuming clean source data and DPIA scope clarity. Week 1 to 4 is discovery, EU AI Act and GDPR classification, and source-data audit. Week 5 to 12 is RAG indexing, agent orchestration, evaluation harness, and human-in-the-loop UX. Week 13 to 16 is integration with Salesforce, ServiceNow, SAP, Talkdesk, OutSystems, or core banking endpoints. Week 17 onward is shadow mode, gradual rollout, and monitoring tuning. Multi-agent enterprise rollouts extend to twenty-eight weeks. We have shipped to this cadence inside DORA-supervised and CNPD-reviewed environments.

Yes. We have delivered technical work-package contributions into Horizon Europe consortia and Portugal 2030 programmes through prime-contractor or research-partner relationships with university groups at the University of Lisbon, IST, and NOVA SBE. Our engagement model on EU-funded projects respects the grant agreement, the consortium agreement, the data management plan, and the open-access publication obligations under Horizon Europe Annotated Model Grant Agreement and Plan S. For Portugal 2030 (PT2030) projects, we align with the relevant SI Operacional programmes through AICEP, IAPMEI, or ANI as the funding instrument requires. We deliver against the prime contractor's project officer reviews and produce the periodic and final technical reports in the format the European Commission expects.