Portfolio
Portfolio
Reference AI cases, documented the way we approach a real project, and the apps we have built for companies such as Deliveroo, Trainline, Vueling and Wallapop.
- Each case, from the business problem to how the result is measured.
- Solutions on Microsoft 365, Azure and Power Platform, or custom-built.
- More than 15 years building software and apps.
Reference AI cases
How we approach an AI project, end to end
These are illustrative cases based on a fictitious distributor, not real clients: they show the solution, how it is rolled out and what is measured. The figures for a real project come from its own pilot.
-
Internal knowledge assistant
The team wastes time hunting for procedures in folders and emails, and internal support answers the same questions over and over.
- Solution: an agent in Teams built with Copilot Studio (or Azure AI Foundry) that answers from SharePoint documents and cites the source.
- Roll-out: a pilot with one department, a prior review of SharePoint permissions and one AI Champion per area.
- What is measured: search time, internal support queries and accuracy on a bank of real questions.
- Key point: the assistant is only as good as the documentation; the pilot also helps uncover outdated documents.
-
Email and request triage
Requests arrive in a shared inbox and someone has to read, classify and forward each one.
- Solution: a Power Automate flow with Azure OpenAI that classifies, extracts the data and drafts a reply.
- Roll-out: routing rules agreed with each team and human review of every draft during the first weeks.
- What is measured: average resolution time, requests routed correctly first time and drafts accepted unchanged.
- Key point: start by suggesting, not sending; autonomy grows when the data justifies it.
-
Invoice and document processing
Finance staff key invoice and delivery note data into the ERP by hand, with errors and month-end delays.
- Solution: Azure AI Document Intelligence and Power Automate read each document, validate the data and send it to the ERP.
- Roll-out: business validations (totals, tax ID, matching order) and a queue for doubtful cases.
- What is measured: hours of manual work, error rate and documents processed without intervention.
- Key point: business validations catch what the model misses.
-
Natural language data questions
Every new question about sales or margins turns into a request to the data team.
- Solution: Copilot on a Power BI and Fabric semantic model, with the query shown in every answer.
- Roll-out: agreed business definitions, per-person permissions and a bank of questions for validation.
- What is measured: report requests to the data team, answer accuracy and weekly usage.
- Key point: if each area means something different by “sale”, that has to be agreed before connecting the AI.
-
Customer churn prediction
Mid-sized customers buy a little less each time and nobody notices until they have gone.
- Solution: a classical machine learning model on ERP and CRM data, with a weekly list of at-risk customers and the reason.
- Roll-out: the list reaches the sales team's CRM, and they decide the action for each customer.
- What is measured: churn among flagged customers compared with a control group.
- Key point: without a control group there is no way to tell whether retention is working.
-
HR assistant with governance
HR spends much of its time answering questions about holidays, leave and payslips.
- Solution: an agent in Teams built with Copilot Studio that answers from the policies and passes anything involving a decision to a person.
- Roll-out: written limits to stay outside the AI Act's high-risk category (no CV screening or assessment of people) and a data protection review.
- What is measured: questions answered, handovers to HR and staff satisfaction.
- Key point: decide and document from the outset what the assistant will not do.
Method
How we document every project
The same structure for every case, so they can be compared and the result can be checked.
-
Business problem
What it costs today, whom it affects and how often, in the language of the area rather than of the technology.
Measured baseline
-
Solution and architecture
What the AI does, with which data and systems, what a person decides and why that technology was chosen.
Solution outline
-
Roll-out
Pilot, change management, training and AI Champions in the area.
Pilot in real use
-
Results
Before-and-after indicators (time, errors, usage, satisfaction) and return against cost.
Before-and-after comparison
-
Lessons learned
What worked, what did not, which risks came up and what we would do differently.
Recommendations for scaling up
App development
Apps we have built
These are some of the main projects we have worked on at Neural Soft:
Deliveroo Android App
Deliveroo is a food delivery service that runs through a mobile app, commonly known as the “Deliveroo app”. It lets customers order food from a wide range of local restaurants and have it delivered to their door. The app uses the customer’s location to show a list of nearby restaurants that partner with Deliveroo. Customers can browse menus, choose their dishes and place an order in the app. They can also track their order in real time and get updates on the status of their delivery.
Trainline Android App
An app for buying bus tickets, European rail journeys and cheap fares. It lets users:
- Buy tickets for all their train and bus journeys in one place.
- Compare train and bus options from 260 companies and find cheap train and bus tickets.
- Compare train prices and buy train and bus tickets in their preferred currency: USD, GBP, EUR, AUD, CAD, CHF and SEK.
- Pay with Amex, Apple Pay, PayPal and all major credit and debit cards.
- Add railcards and discount cards to get cheaper fares on UK rail services, Eurostar, SNCF, Thalys and Renfe.
- Book in advance or buy tickets on the day, up to 15 minutes before departure.
- Buy tickets on their phone and skip the station queues. Mobile and e-tickets are available on selected routes.
Gymshark Android App
Gymshark is a fitness clothing and accessories brand that sells through the “Gymshark app”. The app is designed to give customers a seamless shopping experience for Gymshark products, along with a range of fitness and lifestyle features. Customers can browse and buy Gymshark products, including leggings, tops, sports bras and accessories. They can also create a profile to save their favourite products, track their order history and get personalised product recommendations. Beyond shopping, the app offers fitness features such as training plans, exercise tutorials and healthy meal ideas. It also has a community side, where users can connect with other fitness enthusiasts, share their progress and take part in challenges.
KBC Android App
KBC Mobile lets users manage their banking and insurance easily and securely: make payments, transfer money or check their balance without a card reader. It also gives access to a wide range of extra services, such as renting a shared bike, buying cinema and event tickets, paying for parking, ordering service vouchers and much more.
Vueling Android App
The Vueling app lets users book cheap flights, choose the fare that best suits their trip and personalise it with premium services. Users can book flights, check in, get their boarding passes, access Vueling Club and check their flight status.
Wallapop Android App
Wallapop is the leading free app for buying and selling second-hand goods, promoting a new, sustainable way of consuming based on the circular economy and fair trade. More than 15 million people use it!
WeFox App
The WeFox app keeps all of a user’s insurance policies on their phone, ready to access any time, anywhere. Every type of policy can be managed entirely online.
Quitify App
Quitify helps users with one of the hardest decisions of their lives: quitting smoking for good. It gives advice on how to quit based on expert guidance from the AECC (Spanish Association Against Cancer) and the WHO. Users can follow their progress over time, see consumption statistics and track what they achieve each day.
Letsbonus App
Letsbonus lets users find group-buying discounts on products, services and travel.
Clients I have worked with
Events gallery
Would you like your project to be the next case?
Tell us which process you want to improve and we will propose a pilot with clear indicators from day one.