Developers¶
RecSys is an auditable recommendation system suite with deterministic ranking and versioned ship/rollback.
Who this is for¶
- Developers / platform engineers integrating the suite
- Application engineers wiring one recommendation surface end-to-end
- Recommendation engineers validating ranking changes
What you will get¶
- The fastest path to a runnable local setup
- A minimal “first integration” checklist (one surface + logging + first eval report)
- Links to the canonical API, contracts, and on-call runbooks
Key terms
- Tenant: a configuration + data isolation boundary (usually one organization).
- Surface: where recommendations are shown (home, PDP, cart, ...).
- Request ID: the join key that ties together responses, exposures, and outcomes.
- Exposure log: what was shown (audit trail + evaluation input).
Quick paths¶
- Quickstart (minimal, ~10 minutes)
- Quickstart (full validation)
Fastest path to a non-emptyPOST /v1/recommend+ an exposure log. - Run end-to-end locally
20–30 min tutorial to run the full loop on your laptop. - Minimum components by goal
Decide DB-only vs artifact/manifest mode and what you need to run. - Integrate the API
Auth, tenancy, contracts, and copy/paste examples. - Integration checklist
One-surface checklist: identifiers, attribution, join-rate, fallbacks, rate limits. - Run evaluation and make ship decisions
Validate logs → produce a report → decide ship/hold/rollback. - API reference
OpenAPI / Swagger UI and examples. - Operations runbooks
The first pages to open when something goes wrong. - Troubleshooting for integrators
Symptom → cause → fix checklist for empty recs, auth issues, and join problems.
Developer ladder (recommended)¶
Follow this path in order:
- Get a non-empty recommendation response + one exposure log:
- Quickstart (minimal, ~10 minutes)
- Integrate one surface in your app (for example:
home_feed): - How-to: integrate recsys-service into an application
- How-to: Integration checklist (one surface)
- Emit exposure logs and outcome logs with the same
request_id: - Data contracts
- Exposure logging and attribution
- Produce your first report and make a ship/hold decision:
- How-to: run evaluation and make ship decisions
- Do one rollback drill (so you trust the lever before you need it):
- Operational reliability and rollback
Integration checklist¶
Use the canonical checklist (with anchors you can share in PRs/issues):
- Integration checklist
- Integration spec (headers, tenant scope, request_id): Integration spec (one surface)
Quick sanity check:
-
request_idis stable across exposure + outcome logs - join-rate is not near-zero
- at least one KPI and one guardrail metric exists
Need guided help?¶
- RecSys Copilot (Custom GPT): Recsys Copilot
Read next¶
- Exposure logging & attribution: Exposure logging and attribution
- Candidate vs ranking: Candidate generation vs ranking
- Ranking reference (signals, knobs, determinism): Ranking & constraints reference
- Verify determinism: Verify determinism
- Verify joinability: Verify joinability (request IDs → outcomes)
- Security, privacy, compliance: Security, privacy, and compliance (overview)