AI & product
DiffCV
An open-source CV optimization platform — a structured Master Profile that feeds per-posting CV generation, with a CRM-style application pipeline. Architecture and dashboard built; the AI engine is still in progress.
Role — Sole author — architecture, feature-driven module boundaries, dashboard.
Work in progress. The five-pillar architecture, data contracts and dashboard are in place; prompt and agent orchestration in the AI engine is not implemented yet.
Highlights
- Feature-driven architecture — five pillars (Master Profile, ingestion, AI engine, rendering, application CRM) isolated under their own folders, communicating only through shared contracts in src/types.
- External infrastructure (Supabase, the AI SDK, PDF libraries) is confined to src/lib, so the feature modules stay swappable and testable.
- Dashboard with aggregate widget modules — tailored resumes, AI next actions — over the application pipeline.
- Rendering targets ATS-safe templates, since a CV that a parser mangles never reaches a human.
Overview
An open-source take on CV optimization: centralise professional history once as a structured Master Profile, then generate a hyper-personalised CV per job posting, keeping every application in a CRM-style pipeline.
The interesting part is the boundary design. Five pillars — profile, ingestion, AI engine, rendering, applications — each live in isolation and talk only through shared contracts (UserProfile, JobDescription, TailoredResume, Application), with Supabase and the AI SDK quarantined in a lib layer. That is what lets the AI engine be swapped or rewritten without touching ingestion or rendering.
Status: the architecture, data contracts and dashboard are built. Prompt and agent orchestration inside the AI engine is the next piece of work, and is not implemented yet.