Versioned prompts and schemas
Each AI-adjacent step records prompt names, prompt versions, schema versions, and validation state.
AI diagnostics
A clean control room for prompts, schemas, and truth guard runs.
This console reframes backend telemetry as a product feature: visible model runs, validation signals, evidence records, and safety decisions in a calm review surface.
Runs are inspectable
Parse, optimize, and truth-guard operations persist status, provider, model, timing, and output metadata.
Unsafe claims are flagged
Rewrite suggestions are classified before display so unsupported resume claims do not slip through silently.
Scores point to real signals
Match reports store requirement-level evidence, similarity signals, and explanation text for review.
Structured outputs stay accountable
AI outputs are stored with validation status, warnings, and repair-ready metadata hooks.
Benchmarks are ready to expand
The evaluation harness can compare parser, matching, and truth-guard behavior across datasets.
Execution trail
From input parsing to guarded rewrite decisions.
The system is structured so each AI operation can be inspected, compared, and improved without losing the product story.
Parse inputs
Resume and job text become schema-first JSON with parse confidence and warnings.
Create match report
The scoring engine writes breakdown JSON, ATS coverage, gaps, and evidence rows.
Generate optimizer draft
The optimizer proposes targeted resume changes based on the parsed job and match report.
Guard every suggestion
Truth-guard runs classify each rewrite as safe, needs review, or unsupported.
Latest session snapshot
- resume_parse · success · local-resume-parser-v1
- job_parse · success · local-job-parser-v1
- resume_optimize · success · local-resume-optimizer-v1
- truth_guard · success · local-truth-guard-v1
What reviewers can verify
- Schema-first extraction contracts
- Stored prompt and model metadata
- Evidence-backed score decomposition
- Explicit unsupported-claim detection