Rosettastein Consulting – home
Track record

Projects that show how we work.

Not a wall of logos, but descriptions with enough substance for you to judge whether we fit your undertaking. Work for institutions is named; commercial clients are described by task and outcome until they release us to name them.

Institutions and research

Project

AOP-XML: an exchange format for toxicological pathways

European Commission, Joint Research Centre (JRC), in the context of an OECD programme

Knowledge about adverse outcome pathways — how an initial molecular interaction leads through intermediate steps to a harmful effect — sat in several databases that could not read each other. We designed the format that connects them.

  • Document architecture based on OECD Harmonised Template 201, agreed with the responsible expert group and with the knowledge base development team
  • Implementation as an XML Schema Definition, version-controlled and ready for handover to a repository under the client's control
  • Java reference application that generates valid documents from domain data, delivered with open source code
  • Validator in three forms — console application, desktop tool and web service — so that data providers can check files before submitting them
  • Export application in C#/.NET Core that converts pathway and chemical data from the existing wiki platform into the new format, running on both Windows and Linux and integrable into the web application
  • Presentation of the format and resolution of open questions at a working session at the US Environmental Protection Agency offices in Durham, North Carolina, with the resulting changes subsequently implemented

Why this is typical of us: the real work was not writing a schema. It was negotiating a version that several teams with different data models could all work with — and documenting it so that it survives without us.

Project

Effectopedia: a platform for quantitative pathway models

OECD, multi-year across several contract phases

An existing scientific software platform needed to move from qualitative descriptions of pathways to quantitative, data-backed relationships — and to become stable enough for a public version 1.0 in the process.

  • Features for quantitative pathways: dose-response curves per test substance, response-response relationships with linear, threshold and experimental approximation, and derivation of those relationships from linked dose-response data
  • Temporal concordance in the relationships between key events — implemented beyond the contracted scope
  • Automated consistency testing: coverage for roughly 2,000 methods identified as priority within a codebase of more than 9,000, plus a crash reporting system for live operation
  • A common authentication service working across all content on the platform
  • Setup and operation of a learning management system for online training and a content management system for the public site
  • Transfer of the OECD-approved pathways from the wiki platform, support for the new exchange format, and a training server on which contributors can rehearse every step including publication without risk
  • Server infrastructure, monitoring and domain management; support through the beta phase with training materials

Why this is typical of us: we did not rebuild. We made a system that had grown organically dependable — tests, authentication, operations and training — so that the scientists could worry about content instead of crashes.

Product development: a sports innovation platform

For several years we have been the technical partner of Powerplay One, an innovation consultancy for the sports business. A book and a collection of paper worksheets have become software in the process — the longest continuous product build in our portfolio.

Project

From a book to an assessment platform

Powerplay One, multi-year, ongoing

The basis is an assessment model for the innovation capability of sports organisations, with ten dimensions and around 84 worksheets. On paper it worked in workshops; as software it had to be weighted, stored, billed and explained.

  • Scoring logic combining weighted quantitative and qualitative questions, resolved into a maturity matrix across three phases
  • Template library with progress tracking and recommendations on which worksheet suits a given result
  • User accounts, a history of successive assessments over time, and PDF export of the result reports
  • Subscription model with a free and a paid tier, payment handling and connections to common project tools
  • A template renderer that produces new worksheets from a configuration file rather than requiring each to be coded — forms, matrices, checklists and canvas layouts from one shared component base
  • Operation on our own infrastructure with reverse proxy, process management and automatically renewed certificates

Why this is typical of us: the hard part was not the interface. It was translating a consulting model into a data structure that can be versioned, billed and later extended — the same task as an exchange format for public authorities, in a different domain.

Project

A knowledge assistant built on the client's own book

Powerplay One

The firm's expertise sat in a book — well written, but not queryable. We turned it into an assistant that answers questions and traces its answers back to passages in the text.

  • Preparation of the book text into just under 500 sensibly cut passages, embedded and stored in a vector database
  • Query and answer service as an API with a connected web interface
  • Migration of the language model and embedding layer from one cloud platform to another without rewriting the application — substitutability was designed in from the start
  • Translation of the German original into English and Arabic through a repeatable processing chain, preserving structure, terminology and special characters
Project

Stress-testing a data architecture

Powerplay One, review of a third party's design

A new product arrived with a database design from a third party and a content inventory of several tens of thousands of entries. We were asked to review it — and rather than read it, we loaded it.

  • Full import of the content into a running PostgreSQL database: over 37,000 records, with every row count reconciled against the handover
  • Found in the process: two mismatches between enumerated types and the actual values, one inconsistency in a special-case path, and a configuration in which database-level access restriction would have been silently disabled
  • A recommendation on the division of responsibility: tenant isolation in the database, content entitlements in the application layer — with the reasoning for why the latter does not map cleanly onto database rules
  • An infrastructure proposal at materially lower running cost than the service originally assumed, with the same data residency in Germany and no dead end: moving later remains a plain database migration
Project

Fan app and fan data for a basketball club

First-division basketball club in Croatia, via Powerplay One

A club wanted to reach its supporters directly rather than through third-party platforms — and to know who they actually are.

  • Mobile app serving club content from a content management system, bilingual Croatian and English, with automatic synchronisation of new posts
  • Live sports data across five competitions — standings, scores, player statistics and match recaps
  • Fan accounts with deliberately optional sign-in: the app is fully usable without one
  • Opt-in push notifications, triggered by publication of new content
  • Audience measurement as first-party collection only, with no third-party trackers
  • Ahead of that, a strategy for a unified fan data foundation: an inventory of the scattered sources — ticketing, newsletter, shop, social channels — and an assessment of realistic options within the club's budget
  • Finally, a launch-readiness review drawing a clear line between finished features and those better deferred on purpose

Industry and mid-sized enterprises

This work is subject to confidentiality agreements. We will name clients on request, to the extent the relevant agreement permits.

Project

Machine analysis of legal contracts

Legal technology company, framework agreement with call-offs

A provider of automated contract review needed a dependable foundation for extracting clauses and granted rights from licensing and cooperation agreements.

  • Assembly and preparation of a corpus of around 200 real licensing and distribution agreements as the basis for training and evaluation
  • Methods for extracting structured information from unstructured contract text
  • Assessment of model quality and comparison against solutions available on the market
  • Delivered as a team under a framework agreement with auditable time and effort records per person

The resulting platform formed part of the basis for a successful funding round of over one million euros.

Project

Practice management system — taking over a failed project

Healthcare provider, two years

A previous team had delivered little beyond interface mock-ups, the data schema contradicted itself, and the codebase had no documentation. We took over and shipped.

  • Assessment of the existing code and written documentation of the contradictions in the handed-over schema and sample data — done first, before a line of new code
  • Rebuild of the database, migrations and application connection; clearing out the artefacts left behind by tooling
  • Domain features: management of practices and opening hours with map integration, favourites for laboratories, diagnoses and radiology, employee and role management, absences, and invitation emails with confirmation tokens
  • Effort estimates per increment and sign-off of designs with the client before implementation

Why this is typical of us: we documented what did not fit together and gave that analysis to the client, even though it was unwelcome. Without that step we would have carried the previous team's mistakes forward.

Project

Risk assessment and simulation for a medical device

Manufacturer of drug delivery systems, multi-year

For a multi-dose eyedropper system it had to be demonstrated what quantities of substance patients are actually exposed to — in a form that could be put before a regulator.

  • Strategy for the exposure-based risk assessment of the product
  • Derivation of permitted limits from the available data
  • Computational modelling and simulation in place of measurements that could not be taken
  • Preparation of the methodology and results for submission to the US Food and Drug Administration

Why this is typical of us: the calculation was not the hard part. Documenting it so that a regulator can follow it was — the same requirement as an exchange format, but with a single, very exacting reader.

Project

Mobile application for a pharmacy network

Pharmacy group in North America

A network of independent pharmacies across several provinces wanted to give customers a single app without giving up the individuality of each location.

  • Store finder by proximity, by city and postal code, and by name, with map view and distance calculation
  • Detail pages per location with owner, opening hours, services offered, and direct call, email and navigation actions
  • Prescription function, store flyers and newsletter sign-up, plus personal favourites
  • Integration with the group's existing branch data
  • Part of a broader telemedicine and prescription management platform, designed for the requirements of Canadian data protection law (PIPEDA)

Our own research

Not every project has a client. We spend part of our time on our own investigations — including ones that end up arguing against going ahead.

Arbitrage across crypto exchanges

A study of price differences between several trading venues: modelling the full cost chain per scenario — deposit, trading and withdrawal fees — alongside the execution and transfer times during which the price advantage decays. Complemented by a calculation on the economics of running our own compute.

Outcome: the nominal spreads between exchanges were real, but after fees and transfer times they could not be captured reliably at the capital scale examined. We did not invest — and we have used the same way of reckoning in business cases for clients since.

Further work

Beyond the projects above, we have contributed to EU research consortia on animal-free safety assessment, taken part in an agency-coordinated international model comparison on endocrine activity, and delivered a web portal for the health communication programme of a US university.

References in detail

We can give further detail on individual projects on request, to the extent the relevant confidentiality agreement permits. For tenders we assemble the usual evidence — reference letters, project descriptions and curricula vitae.

Request references