ReviewOSMediGen strategy portalProduct requirements
The PRD's first build target is MVP Stage 0: a Corpus API + MCP server. It should prove retrieval quality, citation coverage, permissions, and auditability before ReviewOS expands into ingestion, synthesis, and workflow.
Given an approved corpus and a high-stakes question, ReviewOS should return ranked passages with exact source text, document metadata, scores, filters, and permission-safe citations through a read-only API and MCP tool.
User document upload, answer synthesis, contradiction detection, review queues, approvals, memo templates, multimodal search, fine-tuning, broad eDiscovery replacement, and autonomous legal or regulatory judgment.
Release tags
| Tag | Release | Product form |
|---|---|---|
| MVP Stage 0 | Corpus API + MCP server | Read-only retrieval substrate over the existing corpus. |
| MVP Stage 1 | Document ingestion | Controlled corpus maintenance and refresh workflows. |
| MVP Stage 2 | Citation + cross-reference agent | Grounded synthesis, contradiction checks, and evidence packets. |
| MVP Stage 3 | ReviewOS application | Projects, queues, approvals, templates, exports, analytics, and audit. |
| X1-X3 | Extensions | Multimodal search, thinking-model expansion, and recursive retrieval optimization. |
MVP Stage 0 functional requirements
| ID | Requirement | Priority |
|---|---|---|
| FR-001 | Ingest the existing ~50K-document corpus. | Must |
| FR-002 | Parse PDF, DOCX, TXT, and spreadsheet files into text plus structure. | Must |
| FR-003 | Chunk documents into retrievable passages. | Must |
| FR-004 | Extract and store document ID, title, type, source, date, page, and section metadata. | Must |
| FR-005 | Create vector embeddings for semantic retrieval. | Must |
| FR-006 | Maintain a keyword/BM25 index for exact terms, IDs, clauses, dates, and references. | Must |
| FR-007 | Perform hybrid retrieval with configurable weighting. | Must |
| FR-008 | Support metadata filters for type, date range, source, project, confidentiality, study, and compound. | Must |
| FR-009 | Expose a read-only retrieval API. | Must |
| FR-010 | Expose at least one MCP-compatible read-only retrieval tool. | Must |
| FR-011 | Return structured citation-first evidence objects with source text and retrieval metadata. | Should |
| FR-012 | Enforce document-level permissions with no cross-scope citation leakage. | Should |
| FR-013 | Log query, retrieved chunks, latency, errors, tool used, and session. | Must |
| FR-014 | Log every failed parse with reason. | Must |
| FR-015 | Ship a golden-set eval harness with 30-50 known-answer questions. | Must |
| FR-016 | Ship an adversarial prompt set that must fail safely. | Must |
| FR-017 | Capture usefulness rating and structured failure reason when offered. | Should |
| FR-018 | Provide indexing-health status reporting. | Should |
MVP Stage 0 ships with a 30-50 question golden set and adversarial prompt set. Gate on top-5 retrieval accuracy, citation metadata coverage, latency, and safe-fail behavior.
Document-level permissions are enforced before retrieval returns any source text. MVP Stage 0 stays read-only and logs queries, returned chunks, latency, errors, and sessions.
Capture query text, filters, chunks, scores, latency, errors, most queried documents, failed retrievals, usefulness ratings, and opened citations.
Success metrics
| Metric | Baseline | Target | Release |
|---|---|---|---|
| Time to first useful source | Hours-days | <120 sec | MVP Stage 0 |
| Known-answer retrieval accuracy | Uncaptured | >80% top-5 | MVP Stage 0 |
| Citation metadata coverage | Variable | >95% of chunks | MVP Stage 0 |
| Retrieval latency | n/a | <10 sec | MVP Stage 0 |
| Factual-claim citation rate | Uncaptured | >95% | MVP Stage 2 |
| End-to-end cycle time | Days | >20% faster | MVP Stage 3 |
Non-functional requirements