Work

Work and evidence

Explore client delivery, AI workflows and the engineering experience behind Devmojos. Each story identifies the work, its current status and the contribution behind it.

Explore previous project experience
Launched client project

Tribeca Festival Lisboa (SIC)

Website & content platform

A festival website and content platform connecting editorial work with the programme and tickets.

Role — Filipe Contente — Architect, AI delivery lead

Implementation code was produced by AI coding agents under Filipe’s direction. Architecture, decisions, review and release responsibility remained human.

Status — September–October 2026 engagement. The final build-to-live window was unusually short for a site of this scope; that describes one phase, not the complete engagement.

tribecalisboa.sic.pt (opens in a new tab)

Discuss a website or content platform

Public website views from Tribeca Festival Lisboa: homepage, programme and tickets on desktop and mobile.

Running for live publications

AI Editorial Desk

Source analysis and labelled AI-written articles.

Source analysis and article drafting, with source attribution, clear AI labels and configurable editorial controls.

Role — Product direction and architecture

Each publication has its own sources, language and editorial profile. English and European Portuguese are in use today. Source links and AI authorship labels stay with the content. Automatic display is configurable per publication and off by default.

Status — Running for live publications; Devmojos website integration is planned. Discuss how it could fit your editorial process.

Discuss your editorial workflow

AI Editorial Desk: production News API; editorial flow and post examples from the development environment. Devmojos website integration is planned.

Live with test agencies

Brief to Shortlist

Property search for real-estate agents

Turn a client’s brief into a shortlist for a real-estate agent to review, with listing evidence and missing details to confirm.

Role — Product ownership and architecture

The AI asks about ambiguous requirements, and the agent confirms the criteria before searching six Portuguese property portals. It reads candidate listing text and selectively checks photos on kept results. Matched, unmet and unconfirmed criteria remain visible for the agent’s decision.

Status — Working system, live with test agencies; not a production service.

Discuss your shortlist workflow

Brief to Shortlist: brief entry, confirmed search criteria and portal activity. Test-agency screens with development data.

Internal engineering practice

Agentic harness & Devbus

Defined work, recorded handoffs and review

AI agents work with defined responsibilities, shared context and independent checks. Recorded handoffs keep engineering work visible and reviewable.

Role — Filipe Contente — Architect and product owner

AI operator agents organise defined work and coordinate coding agents. The human delivery lead remains responsible for architecture and release decisions. Devbus connects named AI agents and records their messages and delivery status. Four-Eyes Delivery is separate May–September 2026 R&D.

Status — An internal engineering practice for coordinating AI work and supporting human delivery decisions.

Discuss your team's delivery process

Internal engineering method and Devbus demonstration. Illustrative UI with fictional data.

In daily use

Fit-for-Job Model Serving

Model serving for coding agents and business apps

Open-weight language models run on Filipe’s own hardware, with each setup configured and tested for its job.

Role — Filipe Contente — Architect and product owner

Two NVIDIA RTX 5090 GPUs serve very long coding-agent prompts; two NVIDIA DGX Spark machines run a mixture-of-experts model as one engine for Filipe’s business apps. Key changes are tested on graded coding challenges with hidden tests, at the apps’ actual sampling settings. Configuration was written by AI agents under Filipe’s direction.

Status — July 2026–present. In daily use. A speed-up that garbled real output was rolled back.

Discuss your model-serving setup

Illustrative model-serving diagram with fictional data. Internal daily use.

R&D lab · May–September 2026

Four-Eyes Delivery

A deterministic harness for AI coding agents

A command-line tool designed by Filipe, with fixed code running the workflow: specification, planning, implementation, independent review and task sync.

Role — Filipe Contente — Architect and product owner

Filipe defined the agent roles and the rules between them. A finished run is not approval: closure requires an explicit reviewer verdict, and no role approves its own work. An edit counts only when a real file changes. Code was written by AI agents under Filipe’s direction.

Status — R&D lab for Filipe’s current harness, May–September 2026; a different approach.

Discuss your team’s delivery process

Four-Eyes Delivery diagram: a fictional finished run is rejected, repaired and closed only after reviewer approval. R&D lab, May–September 2026.

Closed lab · Simulated only

AI Trading Evidence Lab

Strategy tests and forecast comparisons

A paper-trading research lab that tested rule-based strategies and separately evaluated AI forecasts against market prices.

Role — Filipe Contente — Architect and product owner

Strategies were backtested with fees, then run live on paper. AI models forecast the same prediction-market questions from the same frozen evidence in isolated sandboxes, with a blind test without market prices. Studies were pre-registered and an independent model audited the measurement design. Code was written by AI agents under Filipe’s direction.

Status — September–October 2026. Simulated only. No strategy qualified for real money and no model beat the market price in the observed tests, so the lab was closed.

Discuss your AI evaluation process

Fictional illustration of the AI Trading Evidence Lab’s research workflow. Simulated only; nothing qualified for real money.

Previous projects & customer experience

Understanding a client’s business, users and systems remains central as the way we deliver evolves. Devmojos applies AI within its software studio approach: requirements, design, architecture, implementation, testing and scoped operation. That method builds on delivery experience in healthcare, energy and connected systems.

  • Doktor.de (2021–2025) — digital-healthcare architecture, design and technical/team leadership.
  • Coneva Energy Buddy (2020–2021) — backend provider-data integration and community APIs; lead backend contribution.
  • Fibersail (2022–2023) — wind-turbine measurement analytics and cloud delivery; lead engineering/DevOps within shared team delivery.
  • DEHNrisk (2022) — lightning-risk specification, analysis and reusable domain logic; technical lead for specification.
  • DEHNvironment / DEHNdetect (2024–2025) — two telemetry and customer-API backend services, with shared architecture and delivery.
  • ELIA / 50Hertz (2023) — identity demonstrator for energy contracts, assets and flexibility; technical lead.
  • BABEL (2019) — native image-recognition integration for an OutSystems mobile application; contract integration contribution.
  • GIS / MapStore2 — geospatial platform customisation and backend/frontend integration; technical lead.
  • L’Oréal Paris Portugal (2014–2020 personal engagement) — websites, campaigns and customer journeys; web marketing consulting and development.
  • ClicInvest (2015–2019) — online lending, operations and financial integration; CTO within team/company platform delivery.
  • ICN 2017 Porto (2016–2017 engagement) — conference website and mobile application; company/team delivery contribution.
  • Abbott Digital Passport (2015–2017) — mobile commercial field reporting; software development contribution.
  • Central de Cervejas e Bebidas (2014–2016) — mobile point-of-sale evaluation and reporting; software development contribution.

Work across Devmojos, partner engagements and Filipe’s earlier company and personal experience. Roles and shared contributions vary; dates describe the individual engagements.