IntentGuard engineering portal

What do you want to understand?

Confidence-aware intent classification on BANKING77, documented as it was actually built: one baseline, one transformer, a validation-selected abstention threshold, and a headline result — the baseline wins — reported as measured.

abstain ◂ 0.1684 ▸ accept — the persisted validation threshold, sealed in the artifact

Technical hiring manager

Architecture, engineering decisions, evidence, tests, limitations, and production thinking.

Start technical review

Recruiter

What the project does and what engineering competencies it demonstrates.

Review in 5 minutes

ML / AI engineer

Data, baseline, transformer, abstention, evaluation, and serving — with measured numbers.

Explore ML system

Agentic engineering

FPAT Lite, rules, commands, skills, and governed agent workflows.

Explore FPAT

Deep technical explorer

Navigate the repository as a source-backed engineering knowledge graph.

Open Knowledge Graph

Browse the documentation

Project overviewThe reviewer entry point: lifecycle, measured results, scope.ML system guideHow data, models, threshold, evaluation, and serving fit together.SpecificationThe sole authoritative design source, requirements to test strategy.Evidence & claimsWhat the repository claims and what supports each claim.FPAT & agentsThe governed agent workflow this repository was built with.Repository referenceEvery Markdown file, classified, filterable, and routable.

106 documents370 graph nodes1317 provenanced edges59 claims (28 measured)synced @ fb8581f

Everything on this site is generated from w7-mgfcode/intentguard. GitHub Markdown and Git metadata remain the source of truth; this portal is a projection.