← Selected work

Agent-neutral workflow architecture · Ongoing personal system

Evidence-Backed Career System

I maintain a private, file-backed operating system for evidence, job intake, evaluation, CV generation, and immutable application records.

PythonJSON SchemaLaTeXGitHub ActionsDeterministic validation

01

The challenge

Let multiple agents search, evaluate, and prepare applications without inventing candidate facts or depending on hidden chat memory.

02

The approach

Uses an evidence bank, JSON contracts, exact-source boundaries, deterministic validation, generated CVs, and immutable prepared or submitted application versions.

Facts before prose

Claims carry provenance and confidence before they are shaped into a CV, evaluation, or public case study.

Private truth, public projection

The portfolio receives an allowlisted artifact rather than runtime access to applications, jobs, or evidence sources.

03

The outcome

One private source of truth now drives task-specific job and CV workflows plus a separately sanitized public portfolio projection.

Next case studyReinforcement Learning Racing Agent →