OrientIA: A French AI-driven orientation platform - conversational, sourced, and supervised for post-secondary and career choices.

Team name: OrientAI

Use of AI tools : Claude (Anthropic)

• Purposes: Literature research and part of the prototype development.

• We used Claude where AI is currently most productive, namely literature synthesis and coding assistance, in order to focus our own time on what genuinely requires human judgment: the architectural design, the methodological choices behind the pilot study, the analysis of the results, and the editorial trade-offs in the final report. This usage is consistent with the project's central thesis: generative AI is an effective information-augmentation tool on well-defined tasks, provided it is used as a complement to human judgment rather than a substitute for it.

External feedback & contributions :

• Deo Munduku — Professor at Université Paris Dauphine-PSL — Advisor on the project's launch (methodology).

• Nicol D. — Perspective from an Éducation nationale school psychologist (PsyEN) — Expert feedback: drew the conceptual distinction between information and guidance, and pointed out that career and study guidance is a developmental and reflective process, not merely an information gap to be filled.

• Dr. Christelle Scharff — Academic expert — Expert feedback: regulatory framing under the AI Act ("high-risk" systems classification), risk of biases inherited from source data, and the importance of the "human in the loop" and of the broader ecosystem (parents, teachers, counsellors).

• Lucie Jacquet-Malo — PhD in Mathematics, IA program coordinator at Institut Mines-Télécom — Structured mentoring and expert feedback: validation of the project's framing (equity, sovereignty, AI Act compliance), recommendations on scalability (partnerships with ONISEP and regional authorities, modular architecture, open source), and on the funding model.

• Ndeye Sophie Seck — Participant (other team "Orient") — Comments and discussion: raised concerns about preserving the student's critical thinking and the risk that a "perfect answer" would oversimplify the decision; discussion on the project's public-service stance.

• Hans Zúñiga — Platform contributor — Supportive comment: explicit recommendation that the system should never make final decisions on the student's behalf.

• Vaibhav Kumar — Participant — Substantive feedback: endorsed the information/guidance distinction and the maturity of the non-prescriptive approach.

• Kathleen Somers — Participant — Design questions: initial personality questionnaire versus neutral platform; possibility of suggesting related occupations from a single job profile.

• Nipun Ranchhod Navadia — Participant — Constructive feedback: improvements on scalability while preserving neutrality, long-term governance, and ongoing bias monitoring.

• Anthony Miranda — Participant — Supportive comment: endorsed the neutrality of the ranking mechanism (official labels vs. commercial visibility) and the technical RAG approach.

Initial contribution: OrientIA: Reclaiming AI-Driven Guidance - Towards a Sovereign, Neutral and Equitable System for All

Final contribution: https://drive.google.com/drive/folders/1x0SbFbJl4D0ne2lGMHw0sbsbFZ-k7EjU

Reflection on the process (important)

The initial project rested on two deliverables: a quantitative pilot study testing, at the individual level, the link between the quality of the guidance received and the likelihood of reorienting after entering higher education; and a sovereign RAG prototype drawing on French public data (Parcoursup, ONISEP, ROME 4.0, SecNumEdu labels) with a re-ranking mechanism based on official labels. The feedback received during Phase 2 shaped the proposal along three main axes.

What feedback influenced the work. The most structuring input came from Lucie Jacquet-Malo (PhD in Mathematics, IA program coordinator at Institut Mines-Télécom), who mentored us throughout Phase 2 in depth. Her feedback, delivered in three successive stages, first validated our initial framing (equity, sovereignty, AI Act compliance), then opened the discussion to the concrete conditions of scaling up (institutional partnerships with ONISEP and the regional authorities, a modular architecture allowing independent updates of each data source, and open-sourcing of the code to enable public audit), and finally addressed the funding model (a combination of public subsidies and institutional partnerships). It is this mentoring that allowed the project to move from a technical demonstration to an operationally credible proposal.

Other, more targeted feedback also shaped the proposal. Nicol D. (PsyEN perspective), echoed by Vaibhav Kumar, drew the conceptual distinction between information and guidance: career and study guidance is not merely an information gap to be filled, but a developmental and reflective process. This point was reinforced by Ndeye Sophie Seck, who questioned the risk that a "perfect answer" might short-circuit the personal maturation of the choice, and by Hans Zúñiga, who recommended that the system should never make final decisions on the student's behalf. Dr. Christelle Scharff's expert comment strengthened the regulatory framing (AI Act, "high-risk" classification, biases inherited from source data) and stressed the need to embed the project within a broader ecosystem of parents, teachers, and counsellors. Kathleen Somers raised concrete design questions, and Nipun Ranchhod Navadia pointed to long-term governance and ongoing bias monitoring.

What changes were made as a result. First, following the core of Ms. Jacquet-Malo's recommendations, the scalability and governance strategy was fully developed in the final report: operational anchoring envisaged within the AVENIR(s) program, institutional partnerships with ONISEP and the regional authorities, a modular architecture for source updates, continuous bias monitoring, algorithmic documentation to meet AI Act requirements, and an explicit public-funding model. Second, several anti-prescription safeguards were made explicit: the AI systematically rephrases what it has understood of the student's profile before offering any suggestion; it provides multiple weighted options rather than a single answer; and it closes each exchange with an open question designed to support reflection rather than settle it. Third, the articulation with human counsellors was reinforced: OrientAI is repositioned as a three-module platform, one of which is explicitly a space for booking an appointment with a guidance counsellor, kept separate from the AI conversations to preserve confidentiality.

How this process strengthened the final proposal. The exchanges, allowed the project to evolve from a primarily technical demonstration (sovereign RAG and empirical study) into a mature socio-technical proposal: OrientAI is no longer presented as an answer-providing tool, but as a public infrastructure for supporting reflection, articulated with human counsellors, legally compliant with the AI Act, and equipped with a credible operational pathway toward scaling up. The information / guidance distinction, made central thanks to the experts' feedback, now constitutes the conceptual core of the final proposal.

Video pitch

Please provide a link to your 1-minute video pitch presenting your work: https://drive.google.com/drive/folders/1zZNLPpmVnvCpmPeeAJY-GiIc4VCFRoKP