Challenge
French post-bac applicants have to navigate Parcoursup, the national higher-education admissions platform, where more than 23,000 formations compete for attention against a calendar of fixed and unforgiving deadlines. Prospective students were doing that largely unaided: comparing courses across scattered official pages, tracking wishlist and confirmation dates by hand, and drafting motivation letters with no real sense of what admissions panels look for. The consequence was avoidable stress, missed deadlines, and decisions driven by whatever information happened to surface first rather than by genuine fit. The client wanted a guidance layer that sat alongside the official process in French, for families rather than for counsellors.
French post-bac applicants have to navigate Parcoursup, the national higher-education admissions platform, where more than 23,000 formations compete for attention against a calendar of fixed and unforgiving deadlines. Prospective students were doing that largely unaided: comparing courses across scattered official pages, tracking wishlist and confirmation dates by hand, and drafting motivation letters with no real sense of what admissions panels look for. The consequence was avoidable stress, missed deadlines, and decisions driven by whatever information happened to surface first rather than by genuine fit. The client wanted a guidance layer that sat alongside the official process in French, for families rather than for counsellors.
Solution
We built Parcoursup Zen as a French-language guidance platform covering the whole applicant journey. A profile-matching engine scores a student against the full catalogue of more than 23,000 formations and surfaces the ones that genuinely fit, rather than the ones that happen to rank highest. A deadline tracker turns the Parcoursup calendar into personal, dated reminders. A drafting assistant helps students structure motivation letters, and an orientation chatbot answers the questions families actually ask late at night, when no counsellor is available.
The stack is Next.js on the front end with LangChain and LangGraph orchestrating OpenAI models, and Stripe handling subscriptions. Every AI feature was built to run in production against real applicant data rather than as a demo, which meant designing for wrong answers, partial profiles and the French-language nuance the official documentation is written in.
We built Parcoursup Zen as a French-language guidance platform covering the whole applicant journey. A profile-matching engine scores a student against the full catalogue of more than 23,000 formations and surfaces the ones that genuinely fit, rather than the ones that happen to rank highest. A deadline tracker turns the Parcoursup calendar into personal, dated reminders. A drafting assistant helps students structure motivation letters, and an orientation chatbot answers the questions families actually ask late at night, when no counsellor is available.
The stack is Next.js on the front end with LangChain and LangGraph orchestrating OpenAI models, and Stripe handling subscriptions. Every AI feature was built to run in production against real applicant data rather than as a demo, which meant designing for wrong answers, partial profiles and the French-language nuance the official documentation is written in.
Results
The platform shipped as a working product rather than a prototype, covering course discovery through to confirmed choices. Students get orientation guidance at the moment they need it instead of waiting for a scheduled appointment, and the deadline tracking removes the single most common cause of a lost application. Because the matching engine reads the entire formation catalogue rather than a curated subset, it surfaces credible options a manual search would never reach.
The AI layer runs in production with real usage, which is the bar we hold ourselves to. The build also proved out a pattern we now reuse: LangChain and LangGraph orchestration behind a Next.js product, shipped in two months by a small senior team.
The platform shipped as a working product rather than a prototype, covering course discovery through to confirmed choices. Students get orientation guidance at the moment they need it instead of waiting for a scheduled appointment, and the deadline tracking removes the single most common cause of a lost application. Because the matching engine reads the entire formation catalogue rather than a curated subset, it surfaces credible options a manual search would never reach.
The AI layer runs in production with real usage, which is the bar we hold ourselves to. The build also proved out a pattern we now reuse: LangChain and LangGraph orchestration behind a Next.js product, shipped in two months by a small senior team.
