Engineering Research Assistant
From a published finding to an executed experiment with a measured result, in one sequence. The corpus is 0 real arXiv papers, every identifier resolved against the arXiv API, plus a small set of clearly-labelled synthetic internal engineering notes.
Retrieval, extraction, every statistical test and the backlog are model-free. Only the synthesis paragraph on screen 2 uses a model, and it degrades to extraction.
Licence resolution
There is no such thing as "the arXiv licence". Each submission carries its own, and
the Atom API does not return it — this build resolves every paper individually
through the OAI-PMH arXivRaw interface, at one request every 3.2 seconds
with a descriptive User-Agent, as arXiv's terms of use require. Four different
licences appeared across 0 papers.
| Licence | Papers |
|---|
A PDF is downloaded only when the licence allows the full text into the index. For a link-only paper there is nothing to gain from holding it, and for a no-derivatives paper chunking it would itself be the infringement.
What is in the index, and what is not
Internal engineering notes synthetic
Retrieval has to span internal project reports as well as external papers, because that is the actual enterprise question: does what we measured agree with what was published? These 12 notes are invented for this demonstration. They are marked synthetic in the record, inside the indexed text itself so the marker survives into any prompt, and on every citation in the UI.
| ID | Title | Team | Date | Type |
|---|---|---|---|---|
| ENG-2411-017 | Cross-encoder reranking on the maintenance-manual index: latency measurements | Platform Retrieval | 2024-11-08 | benchmark |
| ENG-2502-004 | Reranking did not help on our short factoid queries | Platform Retrieval | 2025-02-19 | experiment-report |
| ENG-2503-011 | Semantic chunking trial: cost was real, the gain was not | Knowledge Systems | 2025-03-27 | experiment-report |
| ENG-2505-023 | Incident: the assistant quoted a torque value from the wrong compressor variant | Field Systems | 2025-05-14 | incident |
| ENG-2506-002 | Why we rejected GraphRAG for the maintenance corpus | Knowledge Systems | 2025-06-03 | design-note |
| ENG-2507-009 | Judge agreement audit on the answer-quality metric | Evaluation | 2025-07-22 | experiment-report |
| ENG-2509-014 | Structure-aware extraction: what it actually bought us | Knowledge Systems | 2025-09-09 | experiment-report |
| ENG-2510-006 | Embedding model selection under the CPU budget | Platform Retrieval | 2025-10-15 | benchmark |
| ENG-2511-021 | Postmortem: our recall number was measured against the wrong gold set | Evaluation | 2025-11-26 | postmortem |
| ENG-2601-008 | Standing constraint set for retrieval work, 2026 | Architecture | 2026-01-12 | design-note |
| ENG-2602-013 | Long-context experiment: cost per answer, not accuracy, was the blocker | Architecture | 2026-02-24 | experiment-report |
| ENG-2604-002 | Retrieval evaluation set: composition and known gaps | Evaluation | 2026-04-07 | design-note |