Product leadership & AI transformation
Founder — TruthX / OpenProof · Paris, France
I turn complex business problems into product priorities, usable workflows and delivery that can be checked. My work connects executive experience, product design and delivery with AI coding agents.
I offer product leadership and AI transformation services: frame a business use case, define the product and user journey, prioritise delivery, coordinate implementation and organise quality checks and adoption.
Paris, France · Services delivered through TruthX.
Discuss a business need · LinkedIn
OpenProof, conceived and led since 2025. I define the user journeys, priorities and source-traceability rules, and lead delivery with Codex / Astra through GitHub changes, reviews, automated tests and continuous integration.
The public demonstration is available. The private workflow remains under qualification. The current structuring kernel is deterministic; integration and evaluation of the delivered multi-agent research prototype remain to be completed.
Explore the public demonstration · French · Product · Research roles and credits
Capabilities: product vision · business requirements · prioritisation · user journeys · delivery with AI agents · traceability · human review · change management.
Product environment I coordinate: Python · FastAPI · SQL / PostgreSQL · GitHub · GitHub Actions · pytest. This describes the delivery environment, not a claim of independent engineering mastery.
1. Explore the fictional map, no sign-in
French · 🇬🇧 English
2. Read contributions and publish an observation
Open the public discussion
The map and discussion are public. No OpenProof account is required to explore; a GitHub account is needed only to post a comment. This exercise gives no access to private cases.
A first lesson from an external review: spending above budget is not automatically an economic loss. In the fictional cofounder case, S06 changes who could approve new expenses. The first public review prompted an update to the written reading guide: check the dates, approvals and documented consequences before concluding. The matching guidance is now visible in the public fictional map. The submitted provenance crosswalk v0.1 is preserved. A v0.2 revised draft now corrects the PROV relation directions, adds explicit derivation, and separates operational, review and reconstruction levels. It is not scientific validation.
Review note 01 — What a source does not prove
Published 22 September 2026 · public fictional case · no real dossier material.
- Claim examined — write the conclusion being tested.
- Sources and exact passages — identify what each source actually records.
- Supported, unsupported or contrary — locate the precise inferential step that fails; do not settle for “more information is needed”.
- Human decision and next evidence — accept, rewrite or reject the conclusion, then name the evidence needed to continue.
- Claim: “The reported €60,000 budget overrun proves that the third founder caused a €60,000 economic loss.”
- Sources: S02 records a €120,000 jointly approved cap; S03 reports €180,000 in spending but says the supporting records still need reconciliation; S04 preserves a late-warning acknowledgement; S06 changes approval rules from 5 May while retaining operational coordination.
- Review: the approved cap and reported spending are supported. Payment, actual economic loss and causal attribution are not established. Approval authority and operational coordination must both remain visible, but neither proves causation.
- Human decision: reject the overstatement. Keep S06 beside the conclusion and ask which transactions were committed or paid, when, under which approval rule, and with what documented consequence.
This example incorporates the rejection threshold made precise in AlekseiUL’s public review. It is a review specification, not a test result for an automated model.
- Source-based review — #44: challenge one formulation and cite the passage that supports or limits it.
- Map readability — #45: show whether a reader can find S02, S03, S04 and the remaining question.
- Technical citation check — #46: separate “the reference resolves” from “the reference supports the conclusion”.
Following is optional. The public fictional map can be explored without an OpenProof account; posting a comment requires GitHub, and private dossiers require a separate invited access. Fifteen minutes is a suggested scope, not a measured completion time.
Request an OpenProof Legal pilot · Application · GitHub organization · Website
I am building OpenProof to help people examine complex documentary cases: recover the context, trace claims to sources and keep conflicting evidence visible.
My work connects experience in transformation and crisis management with product design: who could decide, what was known, what changed and what still needs human review.
Three cofounders built a business together. After their relationship breaks down, one faces a substantial financial claim. Which decisions were shared? Which shortcomings remain personal? Which losses still need evidence?
Explore the fictional case. Its eight excerpts and findings were written for the demonstration; they are not an automated product result.
Follow this profile for small documentary review challenges, source-based corrections and the next contribution to examine. The useful result is a record another person can check, including what challenges the initial interpretation.
- Review one claim: cite a passage, challenge an inference or identify missing evidence.
- Improve readability: help a reader find the source and the remaining question.
- Specify citation checks: describe a valid reference and what should happen when it is invalid.
Contribution scope and rights. Accepted work is credited to the contributor's chosen public name or handle.
The public repository contains a draft RPO format, fictional examples and a local integrity checker. Try the existing integrity exercise.
The full application and TruthX Engine remain private. A matching fingerprint does not prove source authenticity, factual truth or legal validity. The public exercises do not demonstrate automatic document analysis.
I conceived the initial thirteen-module architecture and lead product direction. Under Professor Gaël Dias's academic supervision at GREYC / Université de Caen Normandie, Lucy Martin and Clément Correia-Peltier developed a research prototype and delivered code, a report and a presentation in May 2026.
Integration of this prototype into the current engine remains to be implemented and evaluated. Research contributions and credits.
Discuss an OpenProof pilot · Product leadership & AI transformation · Method and technology
For an initial enquiry, describe the need without sending confidential case files.
♥ Sponsor OpenProof’s public work: support the public integrity checker, reproducible tests and fictional examples with a one-time or monthly contribution. Sponsorship is optional and is not required to follow, try the exercise or contribute.