A typed career pipeline that collects public ATS listings, normalizes and deduplicates them, applies eligibility rules, and isolates optional AI evaluation and notifications.
Job listings arrive through different public ATS formats. Useful automation needs bounded collection, deterministic normalization, evidence-aware evaluation, and duplicate-safe notifications.
Built the FastAPI surface, public ATS collectors, bounded HTTP client, normalization and eligibility services, persistence and migrations, provider boundaries, security checks, tests, and CI.
External listing text is untrusted and inconsistent, while optional AI and notification providers must remain testable without credentials or live calls.
Collectors use documented public ATS endpoints only.
The system does not sign in, bypass CAPTCHA, or submit applications.
OpenAI and Discord require user-provided credentials and are optional.
Collectors target documented public Greenhouse, Lever, and Ashby endpoints through a response-bounded client. Deterministic cleanup, fingerprints, and eligibility rules run before an injectable AI evaluation boundary and deduplicated notification service.
Timeout, redirect, response-size, and retry limits
Deterministic source and cross-source deduplication
Prompt sanitization and evaluation budgets
Cached successful evaluations
Fake providers and blocked unmocked network calls in tests
Secret scanning and sanitized logs
A local, documented backend pipeline with migrations, dry-run flows, demo evidence, security boundaries, and quality checks. No frontend or hosted deployment is claimed.
Tayyab can add AI to an operational system without making the model an untestable or unsafe center of gravity.
Deterministic filtering should reduce cost and ambiguity before AI evaluation.
External text must be treated as untrusted input.
Provider interfaces make AI and notifications testable without live secrets.