PROJECT 05 / 07 · FINTECH · AI GOVERNANCE
Trust before action.
Trust Layer — Multi-Agent AI Governance for Financial Workflows
Human oversight at the boundary between AI suggestions and financial actions.
01 / THE CONTEXT
The challenge.
How can financial workflows benefit from AI agents without granting those agents unchecked authority over sensitive decisions?
02 / MY ROLE
Work with
purpose.
Contributed to frontend and AI backend development within a four-person hackathon team, working across the MANDAT interface and governance backend. Architecture, multi-agent orchestration, security controls and deterministic validation were collaborative work; the overall prototype is not attributed to one participant.
03 / ENGINEERING NOTES
Contributed to the frontend and Python AI backend; the team’s architecture uses LangGraph orchestration to separate AI proposals from permission checks, deterministic validation and authorized execution.
Collaborated on security controls, evidence checks and human approval patterns for payment, compliance and treasury scenarios, with traceable allow, block and require-approval states.
Achieved third place in FINAXYS Build What’s Next, confirmed by the organizer; the team was awarded €200 in AI credits.
04 / HIGH-LEVEL SYSTEM
From idea to structure.
Public-facing conceptual overview — not a disclosure of internal implementation.
05 / THE STACK
The tools
behind the work.
06 / MOTION STUDY
See the concept.
Conceptual workflow visualization — no connected payments or security engine.
Original illustrative concept, not recorded production footage, a real transaction, or an operational system.
MANDAT / TRUST LAYER / PROJECT CAPTURES
The prototype, in view.
An explicitly labeled AI recreation of the award, the original team photograph, and four views of the MANDAT interface. Browser simulations sit alongside a local Python integration panel; the application captures do not establish end-to-end live AI execution.
The MANDAT application screenshots use synthetic demonstration data. No real banking payment, balance or production security certification is represented.
AI-Recreated Award Illustration
An AI-recreated visual of the team’s third-place recognition. Its lettering, artwork and signature are reconstructed; it is not an original document scan. The next image is the original team photograph with the physical certificates.
Source PNG contains OpenAI/GPT and trainedAlgorithmicMedia provenance. This recreation is presented as artwork.AI proposes. Checks constrain. Humans authorize.
FINAXYS Build What’s Next Hackathon 2026 · Third place, team of four. I contributed to frontend and backend development. Architecture, multi-agent orchestration, security controls and deterministic validation were collaborative team work.
The browser workflow illustrates policy decisions. The local Python integration has its own scenario runner and evidence panel. A reachable backend, a counter or a sign-in design does not establish a completed AI workflow, real banking execution or production security.
LinkedIn recap by teammate Stephane Nguene. The organizer’s event page describes the hackathon; it is separate from the team’s account of the result.
FINAXYS Build What’s Next Hackathon 2026 · Third place, team of four. Browser simulations and local Python integration are separate evidence levels. All financial scenarios use synthetic demonstration data. No real payments, genuine balances, certified security product or production banking platform are claimed.