Academic Literature & R&D Synthesis - Patent & R&D Operations

What this group does

Academic Literature & R&D Synthesis grounds R&D work in the actual published state of the art: it curates papers, synthesises them into usable summaries, audits project specs against literature benchmarks, and reviews invention disclosures before they become patent filings. Three of the four screens (Literature Synthesis, R&D Project Spec Audit, Invention Disclosure Review) use the same two-stage hybrid engine as the Prior Art & FTO group — a local-Ollama-only Embedding Engine feeding a writerEngine-routed AI Engine (see Overview).

1. Literature Synthesis

Purpose: Turns a set of academic papers relevant to a target R&D domain into a state-of-the-art summary — consensus level, key methodologies, performance benchmarks, and technology gaps — instead of leaving a researcher to read and cross-reference every paper by hand.

Stage 1 — Embedding Engine

Finds the papers most relevant to the target R&D domain via local Ollama vector search.

Stage 2 — AI Engine

runLiteratureSynthesis() classifies the papers’ overall consensus level — strong agreement / contradictory findings / emerging tech — and produces a state-of-the-art summary plus lists of key methodologies, performance benchmarks, and technology gaps.

Ledger columns

Synthesis ID (SYN-2026-XXXX), target R&D domain, core references count, consensus level, state-of-the-art summary, key methodologies, performance benchmarks, technology gaps.

Gauge

Synthesis Coverage Density — how many tracked research verticals have at least one synthesis, as a percentage of all tracked verticals.

Key actions

Run a new synthesis for a target domain (Stage 1 + Stage 2)
Export Synthesis Brief (exportSynthesisBrief()) — persists the record and returns a structured plain-text brief for copy/paste; no PDF pipeline exists yet

2. Research Paper Library

Purpose: The paper catalog that feeds Literature Synthesis and R&D Project Spec Audit — ingest a paper (pasted text or a dropped .txt/.md file, read client-side via the File API), tag it to an R&D domain, and keep its vector index current.

Ledger columns

Paper ID (PPR-2026-XXXX), DOI, title, authors, journal/conference, full content, vector index status (indexed / pending vectorization / failed), linked R&D domain, AI-extracted methodology summary, added date.

Gauge

Vector Indexing Health Meter — fully indexed papers vs. total.

Key actions

Add a paper (paste or drop a .txt/.md file — full PDF/DOCX binary parsing is out of scope here; that needs the server’s dedicated multipart extraction pipeline used by My Knowledge uploads, a separate integration)
Re-index (reindexPaperVector()) — forces a fresh embedding pass via local Ollama; marks the paper failed if no embedding model is configured or the call errors
Extract Metadata (AI Engine, runPaperMetadataExtraction()) — auto-extracts DOI, authors, journal/conference and a methodology summary from raw paper text, routed through the user’s writerEngine choice like every other AI action in this add-on

Note on embeddings

This screen tracks index status only — it doesn’t persist the vector itself, mirroring how the add-on’s embedding service recomputes on demand rather than caching (a deliberate simplification vs. My Knowledge’s dedicated caching shape).

3. R&D Project Spec Audit

Purpose: Audits a project’s technical specification claims against literature benchmarks for scientific feasibility — catching claims that aren’t actually supported by the published literature before a project plan locks them in.

Stage 1 — Embedding Engine

Finds the literature benchmark references most relevant to the spec’s technical claims.

Stage 2 — AI Engine

runSpecAuditAnalysis() scores overall feasibility (0–100), classifies compliance status — approved / conditional gap / non-compliant — lists specific conflicting claims, and writes a feasibility narrative grounded only in the retrieved benchmarks.

Ledger columns

Audit ID (AUD-2026-XXXX), R&D project name, spec version, feasibility score, academic gap count, compliance status, audit date, AI-identified conflicting claims, feasibility narrative, gap-flag log.

Gauge

Spec Feasibility Index — average feasibility score across all audits, alongside the non-compliant count.

Key actions

Run a new audit (Stage 1 + Stage 2) for a project spec version
Flag Spec Gap (flagSpecGap()) — appends a gap note to the audit’s log; automatically demotes an “approved” audit to “conditional gap” (an approved spec can’t stay approved once a gap is explicitly flagged; a non-compliant audit stays non-compliant)

4. Invention Disclosure Review

Purpose: The formal intake point for a new invention — a researcher discloses what they built, the screen scores its novelty and commercial readiness against prior art, and an approved disclosure becomes a real patent filing.

Stage 1 — Embedding Engine

Retrieves prior art / internal R&D records relevant to the disclosed invention.

Stage 2 — AI Engine

runDisclosureAssessment() scores novelty (0–100) relative to the found references, rates commercial readiness — high / medium / low — identifies the specific inventive steps that distinguish the invention, suggests 1–3 target filing jurisdictions, and writes an assessment narrative.

Ledger columns

Disclosure ID (IDF-2026-XXXX), invention title, lead inventor, target domain, disclosure text, AI novelty score, commercial readiness rating, review status (submitted / under AI assessment / approved for patent filing / rejected), AI-identified inventive steps, AI-suggested target jurisdictions, assessment narrative, linked patent id (once promoted).

Gauge

Patentability Index — disclosures with a novelty score of 70+ vs. total.

Key actions

Submit a new disclosure and run Stage 1 + Stage 2 assessment
Promote to Patent Filing (promoteToPatentApplication()) — only available once approved; creates a genuine new record in the Patent Portfolio Catalog and links it back — a real cross-screen conversion, not a stub. See Screen Workflows for the full flow.

Pages in this section

Frequently Asked Questions

Can I upload a PDF to the Research Paper Library?

Not directly — ingestion accepts pasted text or a dropped .txt/.md file today. Full PDF/DOCX parsing would need the server’s dedicated multipart extraction pipeline (the one My Knowledge uses), which is a separate integration.

What happens if an approved spec audit later gets a gap flagged?

It automatically drops from “approved” to “conditional gap” — an approved status can’t coexist with an unresolved flag.

Can a disclosure be promoted before it’s approved?

No — Promote to Patent Filing is only meaningful once the review status reaches “approved for patent filing”, and a disclosure can only be promoted once (guarded by linkedPatentId).

Does Research Paper Library store the actual embedding vectors?

No — it tracks index status only and recomputes embeddings on demand, rather than persisting and caching vectors.