Product Designer & Researcher1 Designer - 2 DevsAI knowledge assistant1.5-day sprint
A mobile-first AI concept for Hirth Engines: engineers' hands-on expertise flows into one central system during everyday work — without heavy documentation, without interrupting the workflow — and can be retrieved just as dynamically.
When a fault occurs at the test bench, engineers lose hours re-deriving fixes colleagues already solved. The business cost is real: delayed projects, repeated troubleshooting, and risk when expert knowledge leaves.
The core insight from the interview: engineers avoid documentation not out of laziness, but because every existing tool adds friction to an already time-pressured job. New login flows, mandatory forms, switching to a desktop app — each one is a reason not to.
So the design goal inverted: not "make them document more," but remove the documentation step entirely — collect spoken explanations passively and feed them into a hypothesis-driven retrieval system.
"Documentation is mostly avoided by the guys. They try to get around it because it costs time."— Field engineer, Hirth (stakeholder interview)
How might we let engineers capture and retrieve expert knowledge without adding a single step to how they already work?
A 60-minute semi-structured stakeholder interview with a field engineer, mapped to empathy-map quadrants; verbatim quotes extracted for clustering.
Empathy map, point-of-view statement and 12 Jobs-to-be-Done — each grounded in interview signal, assumptions flagged where extrapolated.
How-Might-We questions, feature mapping and explicit scope cuts — kept only what an interview quote or a JTBD could justify.
Mobile-first, AI-generated screens for four core screens — Problem Input, Hypothesis List, Guided Troubleshooting, Quick Capture — refined directly and built out with the team into a working prototype, usability-tested with the lead engineer across two rounds. A lo-fi step wasn't feasible within the 1.5-day hackathon; UX and development worked asynchronously in parallel.
UX artifacts translated into a narrative a non-technical executive audience could evaluate — UX ROI framed as an expert-retention strategy.
Deliverables: interview transcript & empathy map · point-of-view statement · 12 Jobs-to-be-Done · How-Might-We → mechanism mapping · product requirements doc with a success-metrics framework · 4-screen concept · two usability-test rounds on the working prototype · pitch narrative.
Verbatim quotes mapped to the four quadrants — no formal coding scheme, deliberately kept lightweight for a single interview (N=1).
Source: Hirth Engines stakeholder interview · empathy map & pains/gains synthesised from the full research artifact.
→ View full empathy map (PDF)How might we let engineers capture and retrieve expert knowledge without adding a single step to how they already work?
Every job is anchored to a verbatim interview quote, translated into the JTBD format "When … I want to … so that …".
Situation"When a fault occurs that I don't yet recognise — for example scratches on the piston —"
Motivation"I want to see a prioritised list of possible causes right away, instead of starting from zero,"
Outcome"so that I can test the most likely hypothesis directly, instead of losing hours re-deriving it."
Situation"When I'm explaining to a colleague right now how I solved a problem,"
Motivation"I want to be able to record that in the moment, without writing anything down separately,"
Outcome"so that I document it without interrupting my workflow or losing time."
Click to expand — each job in full Situation / Motivation / Outcome format. Grouped into 5 themes: Problem-solving (3) · Information access (3) · Knowledge capture (2) · Trust & validation (2) · Access & adoption (2).
Situation"When a fault occurs that I don't yet recognise — for example scratches on the piston —"
Motivation"I want to see a prioritised list of possible causes right away, instead of starting from zero,"
Outcome"so that I can test the most likely hypothesis directly, instead of losing hours re-deriving it."
Situation"When I ask the system for a solution,"
Motivation"I want to state right away what I've already tried or ruled out,"
Outcome"so that I don't get suggestions I've already tested, and save time."
Situation"When I've settled on a likely cause,"
Motivation"I want a guided step-by-step walkthrough instead of a whole manual,"
Outcome"so that I proceed in a structured way, without reading through irrelevant pages."
Situation"When I get a first answer from the system,"
Motivation"I want to see a short preview before committing to the full depth,"
Outcome"so that I can quickly decide if it's even relevant, without reading long texts."
Situation"When I'm pursuing a cause,"
Motivation"I want linked manuals, diagrams and reports in one place,"
Outcome"so that I stop switching between books, folders and separate AI tools."
Situation"When I have a visible defect in front of me,"
Motivation"I want to upload a photo and get a visual assessment,"
Outcome"so that I communicate the problem faster than describing it in words."
Situation"When I'm explaining to a colleague right now how I solved a problem,"
Motivation"I want to be able to record that in the moment, without writing anything down separately,"
Outcome"so that I document it without interrupting my workflow or losing time."
Situation"When I have a summary slide after troubleshooting,"
Motivation"I want to simply upload it and have the system parse it,"
Outcome"so that I contribute knowledge without manually retyping content."
Situation"When the system gives me an answer,"
Motivation"I want to see who it came from and what it's based on,"
Outcome"so that I can judge it and, if in doubt, clarify directly with the source."
Situation"When I find incorrect or outdated information in the system,"
Motivation"I want to be able to flag and correct it, traceable back to the source,"
Outcome"so that the knowledge base stays reliable for the whole team."
Situation"When I'm at the test bench or out in the field,"
Motivation"I want to use the system from my phone, without detours,"
Outcome"so that I get help without having to go back to a desk."
Situation"When I describe my problem to the system,"
Motivation"I want to be able to do that in full sentences, not just keywords,"
Outcome"so that I convey nuances the system would otherwise miss."
Every kept feature traces back to an HMW and a JTBD.
| Feature | HMW | JTBD |
|---|---|---|
| Voice capture + NLP | Speak & structure a solution | 07 · Voice note on the spot |
| Hypothesis cards | Search like a colleague | 01 · Prioritised list of causes |
| Passive capture | Documentation without documenting | 07 · Voice note on the spot |
| Reliability tiers | Signal confidence | 09 · Source & author per answer |
UX Design built two clickable no-code prototypes (Base44, Replit) for feature exploration and inspirational execution; the dev team developed a working MVP in parallel, focused on technical feasibility — asynchronous rather than sequential.
Focus: technical functionality and feasibility.
Focus: feature exploration and inspirational execution, independent of technical constraints.
Focus: alternative interaction patterns and navigation structure.
The pitch didn't sell UX as aesthetics but as risk reduction: every feature was traced directly back to an interview quote, every cost decision weighed directly against the named pain threshold. For a non-technical executive audience, the product advantage was translated into business risk — knowledge loss on staff turnover, delayed projects, repeated troubleshooting — rather than interface details.
The moat isn't the AI. It's the expert-correction flywheel and the zero-friction capture loop that competitors can't easily copy.
Voice, photo or text — in the moment of solving, with no form.
Auto-tagging and NLP classification organise the input.
Hypothesis ranking by reliability tier for the next query.
Engineer marks a step done / failed / partial — the human stays in the loop.
Corrections flow back into the model — the flywheel spins faster.
Engineers reason forward from causes, not backward from keywords.
Removes the documentation step entirely.
Trust comes from traceable, validated sources.
Direct response to the €20,000/year cost ceiling named in the interview — otherwise the project dies.
No interview signal, no supporting JTBD.
Adds friction to the capture flow.
Originally deferred from the focused 4-screen concept: low mobile utility, no validated job for the primary persona. Room for it turned up within the 1.5-day timeframe after all — built and tested with the lead engineer. Still secondary for the mobile core, but part of the working prototype.
Born in a hackathon and usability-tested across two rounds on the working prototype — here's what's evidenced and what's still an assumption.
Research-output ratio: 1 interview → 12 JTBD → 4 screens → working prototype. Targets (< 2 min to first hypothesis, < 30 s capture, 20–40% faster resolution within 6 months) are pitch projections, flagged [assumption] until a pilot study.
The 1.5-day concept has become an active initiative.
In highly technical contexts, the designer's job isn't to invent features — it's to remove the ones that make experts' lives harder. Every kept decision here traces back to an interview quote; every cut, to the absence of one.