← Pierre Bichon

Sourcing Intake

A personal prototype for triaging early-career structuring profiles, built around one question: what does a fair shortlist look like when you only use evidence?

Personal project · Private prototype · 2026

Sourcing Intake: a dense review table on the left, a profile detail pane on the right
Demo dataset. Apart from my own profile, every name and firm is fictional.

Why I built it

Finding good early-career people in equity derivatives structuring is mostly manual. Profiles are scattered, titles differ between firms, and "structurer" can mean five different jobs. I wanted to see how much of the first pass (sorting, not deciding) a small tool could do cleanly, and what rules it needs so it never does the wrong thing.

What it does

  • Gathers publicly available professional information (role, product area, location, stated experience) into one reviewable list.
  • Enriches each profile with plain, explainable fields: role family, asset class, product type, seniority, career stage, city, and how confident the classification is.
  • Places each profile in one of three bands (target, monitor, low) that a human reviews.
  • Keeps review notes and decisions locally, with an audit trail.

The rules come first

The most important part of the project isn't the code, it's the policy written into it:

  • Evidence only. Role, product, location and public professional evidence. Nothing else.
  • No protected attributes. The tool never classifies or scores on them.
  • Age is never inferred. Not from names, photos, graduation dates or career length. A class year is shown only when someone states it publicly, and it never affects a score.
  • No automated outreach. The tool sends nothing to anyone. Every decision is a human one.

How it's built

A deliberately small stack: a Node.js server, a plain HTML, CSS and JavaScript front end, JSON storage with atomic writes and an append-only audit log. An enrichment module derives each field with rules that can be read and tested. Dense sortable table on desktop, searchable cards on mobile. Unit and end-to-end tests run on a copy of the data.

What I learned

  • Taxonomy is the hard part. Deciding what counts as "structuring" took longer than the UI.
  • A confidence bucket is worth more than a score. It tells you when not to trust the tool.
  • Writing the fairness rules before the features made the design simpler, not harder.

Status

Private prototype for personal use. It isn't a product or a service, and it isn't available publicly.