Ulysse Ha. Trained as an , now building in .
Anthropologist
A qualitative, ethnographic researcher trained in Anthropology (MSc) and Sociology (BSc).
It all began with a heuristic in Anthropology. In today’s world, how are worldviews and behaviours reshaped by our daily interaction with different technological systems? Looking at the intersection of human and digital showed me how behavioural change is induced and sustained by a change in perception of reality. It is still my anchor.
What we create changes who we are.
Taken from Recursivity and Contingency (Yuk Hui, 2019), where I first met recursivity and ontogenesis: the beginning of an evolving ontology, an iteration that is dynamic and recursive.
- 2019–2020MSc Digital Anthropology, UCLWith distinction and a place on the Dean’s list. Talked to a lot of people and wrote about them.
- 2015–2019BSc Sociology, University of BathRead a lot of text.
AI products (Generic mocks used for demo)
-
Find simpler ways to buy and make products
Global rolloutDKK 10m in savings
Help procurement teams see which product variations may cost more than they are worth.
- User
- For procurement teams. The decision owner decides, after Commercial, Procurement and Finance checks.
- What I did
- Created the local proof-of-concept workflow using a local AI development environment.
- Built the SKU profitability and complexity decision flow.
- Connected opportunity discovery to team alerts, review material and human decision gates.
- Shaped the scale path from packaging optimisation to P&L optimisation.
- Board-level spotlight: selected headline for Board of Directors presentation.
- Results
- DKK 2.5m realised in proof of concept · DKK 10m target for 2026 · multi-region expansion · team of 5
- Limits
- The opportunity range is for review, not realised savings. Assignments request checks; they don’t approve the estimate.
-
Help factory teams solve stoppages
Global rollout+1.5% factory efficiency
Give operators troubleshooting guidance that fits the problem in front of them.
- User
- For factory operators. The operator decides, and keeps observation, safe action and escalation.
- What I did
- Built document parsing, canonical knowledge and metadata foundations.
- Designed retrieval, agent integration and approved-source behaviour.
- Created the operator feedback-to-fix workflow and action-pack discipline.
- Automated hill-climbing based on the golden test set.
- Results
- +1.5% overall equipment effectiveness (OEE), target · team of 4
- Limits
- It doesn’t diagnose, approve a restart or replace the site’s escalation process.
-
Make financial reports easier to question
Internal beta95% answer accuracy
Help executives get concise answers with a clear link back to the source.
- User
- For finance leaders. Finance and leadership decide; answers come only from approved reporting packs.
- What I did
- Built the evidence-first answer contract across period, scope, metric and comparison basis.
- Defined source precedence, traceability and explicit partial or not-found outcomes.
- Turned Finance questions into a blind benchmark and repeatable regression loop.
- Automated agent-output improvement from 80% to 95%.
- Results
- Agent output improved from 80% to 95% through automation · in user acceptance testing · team of 3
- Limits
- It doesn’t replace Finance judgement.
-
Plan promotions with a clearer tradeoff
PrototypeDKK 1bn promotion spend, 8 markets
Let commercial teams compare promotion choices before committing their budget.
- User
- For sales and commercial managers. The commercial manager who owns the approval decides.
- What I did
- Designed a common promotion scenario and benefit-calculation method.
- Simulated historical analogue matching, response estimation and ROI comparison.
- Kept model confidence, spend, contribution and assumptions visible.
- Converted Excel-and-email approval into a traceable proposed-to-approved flow.
- Results
- Designed to manage DKK 1bn annual promotion spend · eight markets · proof of concept
- Limits
- Commercial and Finance owners approve every spend decision.
-
Fit products onto pallets with less waste
Internal beta16 products under 5 min per plan
Replace manual packing calculations with a repeatable planning tool.
- User
- For packaging and logistics specialists. A packaging and logistics review decides, choosing from ranked options.
- What I did
- Built the manual-input solver and ranked visual pattern output.
- Converted domain rules—rotation, perimeter, overhang, weight and interlock—into visible checks.
- Used expert-observed failure cases to prioritise engine accuracy over polish.
- Framed the tool for both issue resolution and an upstream NPD quality gate.
- Results
- 16-product pilot (16 SKUs) · target under 5 min to enter a new case · minimum viable product · team of 1
- Limits
- Engineering and operational validation remain required.
-
Find value in spare parts sitting idle
PrototypeDKK 200m+ idle stock found
Spot inventory worth reviewing before teams buy more parts.
- User
- For maintenance inventory reviewers. Maintenance and Finance decide.
- What I did
- Built a site-level priority queue instead of an inflated global opportunity total.
- Separated old-stock, clean cross-site matches and master-data cleanup paths.
- Made calculation definitions and candidate evidence visible at decision time.
- Kept machine criticality, lead time, substitutability and released value as human validation gates.
- Results
- More than DKK 200m in idle stock identified for review, not realised savings · one major business unit
- Limits
- A candidate is not a disposal, transfer or savings decision.
-
Make asset records usable again
Prototype
Help teams see which equipment records need attention and who should resolve them.
- User
- For engineering lifecycle reviewers. Site engineering decides.
- What I did
- Built country comparison, readiness gates, evidence filters and a review queue.
- Made the criticality, lifecycle and project-linkage rule inspectable.
- Separated missing plan linkage from proof that mitigation is absent.
- Added a filtered CSV handoff and a remediation-first path for unready data.
- Results
- No figures published yet.
- Limits
- Blank means unknown. A review flag is an investigation cue, not proof of risk.
The way to Copenhagen
Working across Europe and Asia.