A high-tech startup snags a hefty EU grant. And forgets the ethics assessment. The extremely tight timeframe spurred them to ingenuity.
A satellite-imaging startup landed major EU funding – and then discovered, days before the deadline, that they’d missed the mandatory ethics and compliance assessment. Their documentation existed, but was scattered across the company in dozens of formats with no clear structure.
I built a hybrid process – manual review, LLM-assisted sorting and clustering, and a targeted questionnaire – that turned that mess into a complete, defensible ethics report in a matter of working days. No shortcuts on rigor, and no outsourcing of responsibility to “AI”: the analysis was machine-assisted, but every judgment call stayed human. Plus: The complete process was designed to ensure absolute data safety.
The result: a report that went well beyond the expected compliance minimum, held up to scrutiny from the European Commission, and let the client keep their funding on track – on time and on budget.
Client
Deep Tech, Satellite Imagery / Earth Observation
Year
2024
Scope
Ethics/Compliance Assessment for EU Horizon funding body (European Commission)
Read full story
Expertise Used
Digital & Tech Ethics
EU Compliance
AI & LLMs
Prompt Engineering
Coding
GDPR, Data Safety
IT & Data Risk
Reporting
The Problem: Great at Reading Images,Blind to Their Own Paper Trail
The client is a highly successful startup that processes commercial satellite imagery – photo, infrared, radar – using self-learning algorithms, giving infrastructure operators a real edge: spotting overgrowth along railway lines before it becomes a problem, tracking supply chains for sensitive materials, monitoring seismic activity near pipelines in real time.
They’re experts at finding patterns in other people’s data. Their own company’s data was a different story.
As part of their EU funding (EIC Horizon / EIC Accelerator), the European Commission requires an ethics advisory review. The client had missed it – and when they tried to catch up, they hit a wall. The requirements themselves were vague, and the internal documentation needed to answer them was scattered across the organization: technically “everything was documented somewhere,” but in wildly inconsistent formats, with no way to tell what was relevant or where to look.
The real task wasn’t just finding answers – it was figuring out which questions even needed to be asked. A needle-in-several-haystacks problem.
Naturally, someone suggested there must be an AI tool for this by now.
Nooope. There isn’t. No current model can look at a pile of documents with unclear relevance and unclear content and derive the specific, forward-looking ethics questions a case like this actually needs. And there’s a second, harder problem: an AI can’t hold responsibility for the answers. A human has to.
Pantone® 532 C
C71 M65 Y64 K72
#232323
Pantone® 1795 C
C19 M90 Y79 K9
#a50834
Pantone® 128 C
C3 M14 Y76 K0
#f9d55c
Pantone® 663 C
C02 M01 Y01 K00
#f7f7f7
The Solution: Human Judgment, Machine-Scale Sorting
The answer was a hybrid process – pairing human creativity and accountability with an LLM’s ability to sort,
cluster, and make sense of a large, messy, semantic dataset.
⸺ Collection
Working with a project assistant on the client's team, we pulled together everything that might be relevant: exported intranet pages, saved web pages, slide decks, conference cheat sheets, chat logs, brochures, technical documentation, source code comments, user feedback, plus the company's formal governance and compliance material. Printed out, it would have run to several thousand pages – far too much to review manually in the time available.
⸺ Structuring
Rather than reading everything linearly, the material was clustered by topic and theme using LLM-assisted analysis, surfacing the recurring issues, gaps, and edge cases buried across formats – and making it possible to see, for the first time, what the documentation actually did and didn't cover.
⸺ Targeted prompting
Working from those clusters, we built a structured set of prompts to interrogate the material systematically – pulling out the specific facts, admissions, and open questions needed to answer the Commission's actual requirements, rather than generic ethics boilerplate.
⸺ Follow-the-thread review
Every AI-assisted finding was checked, cross-referenced, and chased down manually – playing detective on inconsistencies, gaps, and the things left unsaid. This is where the human judgment did the real work: knowing which loose thread was worth pulling.
⸺ Report
The findings were consolidated into a clear, structured ethics and compliance report addressed to the European Commission – one that covered the required GDPR groundwork and went noticeably further, directly engaging with the substantive ethical questions raised by the company's technology.
The Result
Given the timeframe, the depth of the final report caught people off guard – including, in a good way, the European Commission's own reviewer.
A complete, defensible ethics assessment delivered within a week.
Exceeded minimum requirements, addressing substantive ethical questions rather than just GDPR checkboxes.
On budget, in scope, with the client’s funding and project status preserved.
The client’s own ethics officer went from panicked to reassured – and the innovation project moved cleanly into regular operations.
The approach worked well enough that it’s becoming a standalone offering:
Ethics as a Service, launched in autumn 2025.
