2026 – ongoing · Remote
Candidature — an agent for preparing and tracking job applications
Entrepreneuriat personnel — Web projects run independently from 2005 to 2011, and the first sites delivered to clients.
Personal project launched in September 2026, alongside the bernaz.fr site whose career database it reads. An agent collects job ads from seven sources, qualifies them against the career record, prepares a CV and a cover letter in the language of the ad — French, English or Spanish — and tracks each application through to closure.
Role held: AI Product Manager — technology stack selection, setup of the AI-assisted development environment (rules written before the first line of code, skills, a dedicated agent), design of the tracking model. The CV and the letter come out as drafts, placed on a Drive to be reworked: no figure appears in them that does not come from the career database, which a check verifies on the document produced, and nothing is sent without human validation. The first screening of ads is done without AI, by a text filter; the agent only steps in on the ads that pass it.
Tracking separates the job ad from the application: an application only exists once the decision to apply is made, and the response rate is computed on applications sent, not on ads viewed. Eleven states, transitions checked in the database, scheduled follow-ups and a daily action list. The code is written with Claude Code; delivery is steered through a Linear backlog of 46 tickets in three weeks.
Achievements
- 3 document languages, matching the job ad
Automatic generation of CV and cover letter drafts
.docx templates in French, English and Spanish, filled from the career database and placed on a Drive to be reworked before anything is sent.
- 11 application states
Application tracking through to closure
Transitions checked in the database; "no response" kept distinct from "rejected"; scheduled follow-ups, a daily action list, and a response rate computed on applications sent only.
Human validation built into the process
The agent prepares, files and records; moving an application from ready to sent is a human step, and no tool sends anything. In the mailbox, the agent can only create drafts.
- 0 figures outside the source data
Anti-embellishment check on every document produced
A model that writes a CV embellishes: a check rereads the document produced and rejects any figure absent from the career database.
- 7 job ad sources
Daily collection and AI-free screening of job ads
Seven sources queried by API every day; only ads never seen before are processed, and an AI-free text filter discards off-topic ones before any qualification.
- 6 skills, 2 of them built for the project
Setup of the agentic development environment
Rules written before the first line of code, an agent dedicated to job applications with its two in-house skills, plus four third-party skills: Word documents, MCP servers, Supabase and Postgres.
- 44 typed MCP tools
Design of the tracking database's MCP tools
Job ads, documents, applications, follow-ups, contacts, Drive and mailbox: every action goes through a typed tool, the only write path to the tracking database.
Skills applied
- Agentic development environment setup
- AI Product Management
- Building skills for AI agents
- Claude Code
- Context engineering
- Designing MCP tools
- Designing specialised AI agents
- Frugal AI
- Human-in-the-loop
- Internationalisation (i18n)
- Linear
- Product ownership
- Relational data modelling
- Supabase
- Technology stack selection
- Third-party API integration