Research
Understanding what people want. Without knowing who they are.
Fixmeapp is a research-based company. Together with KTH Royal Institute of Technology, Aalto University and Eötvös Loránd University we study how a service can understand intent, respect consent and handle data, and we build what we learn into the product.
Ask less.
A good assistant understands a whole sentence and asks one question only when it is worth the person’s time.
Store less.
What a system never keeps, it can never leak. We design for the least data that still does the job.
Prove, don’t collect.
Where something must be known, it should be provable without handing over the data behind it.
01 · Intent understanding
From a vague sentence to the right booking.
People rarely say exactly what they want. They write “something for my hair this week”, change their mind halfway, or add a condition three messages later. This work keeps an explicit, probabilistic picture of what someone most likely wants, decides when a clarifying question is actually worth asking, and lets the language model work from a structured summary of that intent instead of from raw conversations and personal details.
In Fixmeapp
Search in Fixmeapp starts from one sentence, with the service, the place and the time, and the booking assistant asks one question at a time instead of a form.
Ethical AI for Task-Oriented User Intention Inference and Planning under Vague Input
A booking assistant that keeps an explicit, probabilistic picture of what a person most likely wants, decides when one clarifying question is worth asking, and lets the language model see a structured summary of that intent rather than raw conversations or personal identifiers.
KTH Royal Institute of Technology2026 · In progressPrivacy Engineering
Architectural design principles for a centralized data-sharing platform on a social-booking surface: privacy-by-design, GDPR, anonymization, differential privacy, and synthetic data generation.
KTH Royal Institute of Technology2026 · In progress
02 · Consent
Consent you can see, change and take back.
Consent is usually a checkbox at sign-up and then forgotten. This work treats it as something that lasts: decisions you can review in one place, change at any time and trace afterwards. Tested with fifteen participants, the prototype did well on transparency and on avoiding manipulative design. Its clearest finding was that a consent screen cannot fix a system that asks too many questions. The architecture itself has to ask less.
In Fixmeapp
Search history in Fixmeapp has its own switch in Settings, and your data can be exported or deleted from a single page.
User-Centered Consent Flow
Design and prototype of a user-centered consent flow with dynamic consent principles, lifecycle-based design, and a centralized management dashboard. Evaluated with fifteen participants — strong on transparency, with cognitive load identified as the binding scalability constraint.
KTH Royal Institute of Technology2026 · Completed
03 · Ethical data economy
Your data, provable without being exposed.
Two completed degree projects look at the same question from different sides: how a platform can verify facts about a person without storing or revealing the data behind them. One uses predicate-based verification and compares it with raw storage and zero-knowledge proofs. The other builds a Merkle-tree system in which people contribute data, stay its owners and can be rewarded for it, measured against the FACT principles of fairness, accuracy, confidentiality and transparency.
In Fixmeapp
This is the groundwork for what comes next: a way to take part in a data economy on your own terms. Today it is research, not a feature in the app.
Verifiable Data Contribution
A SHA-256 Merkle-tree proof-of-concept for an ethical, incentivized data marketplace — aligned with the FACT principles. Joint thesis at ELTE, co-supervised with Aalto.
Predicate-Based Data Sharing
A WibsonTree-inspired prototype using Ordered Binary Decision Diagrams (OBDD) to verify user attributes without revealing them. Benchmarked against raw storage and a Zero-Knowledge Proof system (Groth16) — proves predicate-based architecture is a practical alternative to both.
Publications are linked here as they appear. We are not taking internship or thesis applications at this time.