We build AI that boosts your company's productivity

• Opportunities and risks Where AI can create value or put the business at risk, process by process.
• Where to invest A business case that compares options, expected benefits, risks and total cost of ownership (TCO) estimates to support the investment decision.
• Roadmap that follows the market A phased plan, reviewed as the market and the technology change.

What you get

• A clear, sourced view of how AI is affecting the economy and your industry, and what that means for the company.
• A portfolio of initiatives ranked by value and feasibility, with the ones that do not pay off now set aside.
• A documented roadmap, with vendor options and dependencies identified, that you can run with us or with another team.

How we work

1. We analyse the economy and the sector using identified sources. We use Business Process Analysis (BPA), or analysis of business processes, to map with the teams how things work today (AS-IS) and identify losses along the value chain.
2. For the priority initiatives, we prepare a business case with options, expected benefits, costs, risks and assumptions; we map which obligations may apply to each system and confirm them with your legal advisers.
3. We define the roadmap, with a budget, performance indicators and baseline values, and design the future process (TO-BE) to prepare the first proof of concept.

Technologies

• BPMN
• PM4Py
• Inspect AI
• MLflow

Examples. We choose tools per project and re-test when new versions are released.

• Order processing Orders read from emails, PDFs or portals and entered into the ERP, with differences flagged, as one stage of the order-to-cash cycle.
• Supplier invoices Invoices checked against the order and the delivery and prepared for posting in the accounts, as one stage of the procure-to-pay cycle.
• Recurring reports Weekly and monthly reports assembled from your systems, with no copying and pasting.

What you get

• One process at a time, tested on real cases and assessed with key performance indicators (KPIs) before and after the change.
• Agents with access only to what they need, and human approval before any action that cannot be undone.
• A record of every agent action and, wherever possible, a way to undo it.

How we work

1. We apply Business Process Analysis (BPA), the analysis of business processes, to map tasks, owners, volumes and exceptions and choose a frequent, checkable step to start with.
2. We run a pilot on real cases, with human approval on irreversible actions and a log of every call to your systems; anything that arrives from outside, such as emails and attachments, is treated as untrusted.
3. We prepare the move to production, with access controls, monitoring and recovery procedures, and only then extend to another process.

Technologies

• Model Context Protocol (MCP)
• Agent2Agent (A2A)
• LangGraph
• Microsoft Agent Framework
• Temporal

Examples. We choose tools per project and re-test when new versions are released.

• Responsible use Recognising errors, protecting data and deciding when to validate an answer or request human review.
• Practice on business processes Exercises using workplace scenarios and approved data to apply learning to the team’s tasks.
• Independence after training Reusable materials, assessment criteria and an application plan to continue learning at work.

What you get

• A programme adapted to each group’s roles and experience.
• Materials, exercises and examples to reuse at work, with responsible-use guidance.
• A learning assessment and a plan for applying skills to the company’s processes.

How we work

1. Identify participants, current skills, priority processes and the rules for accessing tools and data.
2. Prepare sessions by audience, combining explanation, demonstrations and guided practice with approved scenarios and data.
3. Assess skills through exercises and define next steps for applying learning and supporting adoption.

Technologies

• Jupyter
• Python
• TypeScript
• Git
• Inspect AI
• OpenTelemetry

Examples. We choose tools per project and re-test when new versions are released.

• Sorting objects Trials of separating parts by type, size or defect, with computer vision guiding the arm.
• Feasibility in simulation Simulating the task and the workspace to see whether a robot can do it, before buying equipment.
• Desktop arm prototypes Teaching the task by demonstration to a low-cost arm and measuring what works.

What you get

• A feasibility study, in simulation and on a low-cost arm, and a business case before any purchase of industrial equipment.
• If the study is positive, a proof of concept on one task, with a commercial arm and a controlled environment.
• Success measured over many trials, and safety limits that do not depend on the AI.

How we work

1. We simulate the task and the scene to design the control policy before touching hardware.
2. We collect real demonstrations by teleoperation and train by imitation on a low-cost arm.
3. We measure success over many trials and set safety limits independent of the AI; formal safety assessment of the finished machine stays with specialist bodies.

Technologies

• LeRobot
• Vision-language-action (VLA)
• Isaac Sim
• Isaac Lab
• MuJoCo
• ROS 2

Examples. We choose tools per project and re-test when new versions are released.

• Short delivery cycles Frequent releases, demonstrations and improvements based on real use.
• Removing technical blockers Diagnosing data, integration and performance issues that prevent adoption.
• Handover to your team Documentation, practical sessions and handover so your team can run and extend the system.

What you get

• Software built in your environment, with acceptance criteria agreed before work begins.
• Code and documentation delivered within the agreed scope, with rights defined by the contract and licences.
• A team ready to run the system, with monitoring, recovery procedures and clear responsibilities.

How we work

1. We learn how your operation works with your team and agree the problem, access, scope and success criteria.
2. We build in short cycles within your systems, test on real data and review each delivery with users.
3. We prepare monitoring, recovery and handover, then move to production when the criteria are met; hosting can be cloud or on-premises.

Technologies

• Model Context Protocol (MCP)
• FastAPI
• Next.js
• PostgreSQL
• OpenTelemetry
• Langfuse

Examples. We choose tools per project and re-test when new versions are released.

• Assembly and packing checks Checking the contents of boxes and kits to flag possible missing or swapped parts.
• Visual measurement Dimensions, positions and wear measured from images, without touching the part.
• Infrastructure inspection Images from drones or cameras used to detect cracks, corrosion or damage on roofs, pipes and equipment.

What you get

• A proof of concept on your own images, so you know early whether the problem can be solved, before investing in cameras or integration.
• A success criterion defined with you and measured on images from a different day or shift.
• Criteria for accepting results, referring doubtful cases to a person and limiting automatic decisions.

How we work

1. We define with you, in writing, what counts as a defect or a relevant object and how success is measured.
2. We collect and label images of the objects and equipment to be analysed, set the camera framing to avoid capturing people, and measure the result on a separate test set.
3. We start in shadow mode, where the system suggests and a person decides, and we review component licences before handover.

Technologies

• Object detection and segmentation
• Vision-language models (VLM)
• Docling
• PaddleOCR
• NVIDIA DeepStream
• TensorRT

Examples. We choose tools per project and re-test when new versions are released.

• Triage and routing The reason for the call identified and the caller routed to the right person or department, or a callback request taken when nobody is available.
• Confirmations and follow-up Appointment reminders and delivery confirmations made by voice to customers who asked for them.
• CRM records Summaries of requests and next steps in the CRM, limited to the information needed for follow-up.

What you get

• One use case at a time, with an agent that identifies itself as AI at the start of the call.
• Rules for handing over to a person, with a callback request when the team is unavailable.
• Tests with simulated calls and, where authorised, real calls, including noise, interruptions, numbers, names and regional accents.

How we work

1. We pick one use case, such as bookings or frequent questions, and agree with your team when the agent hands the call to a person.
2. We test voice vendors on European Portuguese with simulated calls and, where authorised, recordings of real calls, including regional accents.
3. We connect telephony and decide what information needs to be kept. We map which obligations may apply to recording, to the use of calls in testing and to data retention, and confirm them with your legal advisers.

Technologies

• LiveKit Agents
• Pipecat
• Speech-to-speech
• Streaming ASR / TTS
• WebRTC
• SIP

Examples. We choose tools per project and re-test when new versions are released.

• Questions to your data Questions in plain language about sales, customers or production, with the answer and the query that produced it.
• Anomaly detection Alerts when a value moves out of the normal range, in consumption, payments, returns or machine stoppages.
• Customers and collections Indicators for the team to follow customers showing signs of lower activity, and overdue payments.

What you get

• A frank answer, right at the start, on whether your data is enough for the question; if it is not, we say what is missing.
• A simple baseline that any model has to beat before it is used.
• Forecasts and alerts in the ERP, CRM or dashboard your team already uses, presented in figures that help you decide.

How we work

1. We audit sources, quality and access, and limit the data to what is necessary. Before using an external API, we map which obligations may apply to processing and transferring the data and confirm them with your legal advisers.
2. We test the solution on a defined task, with separate data for evaluation and criteria suited to the use case.
3. We put the result in the system your team uses, monitor data drift and explain forecasts in business units.

Technologies

• DuckDB
• Polars
• dbt
• MLForecast
• NeuralForecast
• LightGBM

Examples. We choose tools per project and re-test when new versions are released.

• Access and data Who can use each system, what data it sees and where that data is stored.
• Incident response A clear procedure for when an agent gets it wrong: detect, stop, fix and learn.
• Policies and change management Internal rules for using AI, plus communication and team training, with follow-up on adoption and on difficulties in daily work.

What you get

• An inventory of the AI in use and its risks. A RACI matrix showing who does the work, who is ultimately accountable, who is consulted and who is informed.
• Monitoring, testing and an incident process that work in practice, not only on paper.
• Records that make audits easier. We map which obligations may apply to each system and confirm them with your legal advisers.

How we work

1. We inventory the AI systems and the data flows and classify the risk of each one.
2. We set owners, acceptance tests, monitoring and an incident process, and assess each vendor: where the data lives, retention and sub-processors.
3. We follow the operation with pinned model versions and regression tests on every change, and review the records regularly.

Technologies

• OpenTelemetry
• Langfuse
• Inspect AI
• NeMo Guardrails
• Open Policy Agent (OPA)

Examples. We choose tools per project and re-test when new versions are released.

• Custom agents Agents designed for the rules and exceptions of your business, when generic solutions fall short.
• Modernising legacy processes Processes built on spreadsheets, paper or old systems moved to a current solution.
• Minimum Viable Product (MVP) A first working version to test the idea with real users before investing more.

What you get

• A short discovery that tells you whether the problem really needs AI; if it does not, we say so.
• A proposal with a defined scope and phased deliverables, with decision criteria (go/no-go) for moving ahead, reshaping or stopping.
• Documented code and a licence review of the open source components before handover.

How we work

1. We run discovery on the problem with the people who live it and write a proposal with a defined scope.
2. We test a proof of concept on your data and decide with you whether the conditions are in place for a pilot with real users.
3. We prepare the move to production once the agreed criteria are met, with documentation and a licence review of the components.

Technologies

• Model Context Protocol (MCP)
• FastAPI
• Next.js
• PostgreSQL
• pgvector

Examples. We choose tools per project and re-test when new versions are released.

• Rules and owners Who approves what, what permissions each agent has and how what it did is recorded.
• Change management Team training and support for adoption, so AI is used day to day and not only tried out.
• Team independence Documentation, training and handover of the operation, so the team can run the agents and knows what it takes to change them.

What you get

• An operating model that defines how people, agents and systems work together, with clear responsibilities, decisions and ways of monitoring progress.
• Every agent with an owner, defined permissions and a record of what it did, and a person approving anything that cannot be undone.
• Data processed in chosen, recorded locations. We map which obligations may apply to each system and confirm them with your legal advisers.

How we work

1. We assess where the company stands today: processes, systems, who decides what, and the AI people already use on their own.
2. We start small: one process, one agent and a clear decision to stop or continue. We test on examples from the company before launch and again whenever the model changes.
3. We widen to other processes and connect the agents to the company's knowledge and tools. We train the people, leave documentation and hand over the operation, so the team can run the agents and knows what it takes to change them. We choose and record where data is processed, and we map which obligations may apply to each system and confirm them with your legal advisers.

Technologies

• Model Context Protocol (MCP)
• Agent2Agent (A2A)
• OpenTelemetry
• OpenFGA
• Qdrant
• Langfuse

Examples. We choose tools per project and re-test when new versions are released.