A narrow, testable task
Start with integration-support questions rather than a general assistant. A bounded corpus makes both correctness and failure easier to measure.
03 / Retrieval · Enterprise integrationsPlanned project
Evidence Desk
An integration answer should come with its evidence.
A proposed support copilot for synthetic integration documentation: permission-aware retrieval, passage-level citations and clear abstention when the evidence is missing.
Brief defined / Implementation aheadPlanned work. This page records the intended problem, design direction and evidence needed for a release. No working product or measured result is claimed.
A problem worth exploring.
Integration teams need to find the right answer in scattered API documentation without confusing a fluent response with an accurate one.
My contribution
Planned personal project. The brief and acceptance criteria are defined; implementation and results are not yet available.
- Use synthetic documentation only.
- Enforce document access before retrieval.
- Keep answers traceable to exact passages.
- Test missing evidence and malicious document instructions.
Follow the flow.
Question
Access filter
Retrieve passages
Cited answer / abstain
Evaluation set
A proposed workflow to test. Architecture and technology choices will be revised as the implementation earns evidence.
What the design needs to prove.
Grounding is an acceptance criterion
The planned release requires a versioned test set, citation verification and documented abstention behavior, including examples where the system fails.
The bar for a release.
- Planned: versioned question set and retrieval measurements.
- Planned: permission-isolation tests and citation support checks.
- Planned: a guided demo with failure examples.
Known limitations
- Concept brief only. No working application, evaluation score or performance result is claimed.
Keep asking better questions.
Build the synthetic corpus and evaluation set before optimizing generation.
Next project / Planned