An async control plane for price and catalog extraction jobs that keeps task history, isolates failures, and recovers interrupted work.
One-off scraping scripts become difficult to operate when pages hang, selectors change, targets throttle requests, or workers stop mid-run.
Designed and built the service architecture, request schemas, extraction registry, task queue, worker flow, persistence model, failure semantics, tests, Docker packaging, and recruiter-facing documentation.
The platform needed predictable ownership of work, explicit retry meaning, per-target errors, and safe recovery without duplicating extracted results.
SQLite is intentionally used for an easy local demo.
Real targets, proxies, cookies, and private selectors are excluded.
The project respects target terms, robots policies, rate limits, and applicable law.
Each target becomes a durable task. Workers lease pending or expired tasks, run one bounded extraction attempt, then persist the result and task state in one transaction. HTTP and browser extractors share a typed registry.
Expired lease recovery for interrupted workers
Idempotent, task-linked result persistence
Explicit retry budgets and terminal dead-task state
Bounded HTTP and browser concurrency
Structured JSON run, lease, retry, and terminal logs
Safe in-place SQLite upgrade path
A locally runnable extraction service with inspectable run history, failure replay behavior, quality gates, CI, Docker verification, and clear operational boundaries. No hosted production deployment is claimed.
Tayyab can turn fragile browser scripts into a maintainable extraction system with explicit failure and recovery behavior.
Retry semantics should be named in total attempts, not inferred from an ambiguous budget.
Recovery is only useful when result persistence is idempotent.
A small operational API can make extraction failures understandable without a full dashboard.