Marconi

Marconi is a patent and intellectual-property research workstation for organizing invention context, source documents, claims, patent families, structures, evidence, and human review. Start an investigation with a clear invention description and intended activity. Record target, jurisdictions, relevant dates, and search scope before interpreting a result set.

The training investigation on first launch contains fictional DEMO- records and is clearly illustrative. Use it to learn the interface only. New investigations start empty, and only real source records that you import or retrieve should inform your own landscape review.

Import patent metadata from CSV, investigation JSON, private text, or PDF. You can retrieve public documents by identifier through configured services; EPO OPS search, family, and claims retrieval require your own credentials and quota. Inspect the source and provenance before adding a document to the investigation.

Record the source universe and collection date for every landscape. The same query over a different database, date window, jurisdiction set, or family policy may produce a different map. Family deduplication helps organize records but does not make separate filings legally interchangeable.

Use Boolean AND, OR, NOT, phrase queries, field restrictions, date filters, jurisdiction filters, and family views to navigate the corpus. Lexical ranking can help prioritize reading, but Marconi's semantic mode is an explicitly labeled lexical approximation; no embedding model is bundled.

Open the document reader before drawing a conclusion from a search hit. Inspect the original source text, claim dependencies, definitions, Markush language, ranges, and cited context. Rule-based segmentation can help organize claims but should not be treated as legal interpretation.

If the investigation includes chemical structures, validate candidate SMILES or MOL files locally with RDKit. Marconi supports exact, SMARTS substructure, supplied scaffold, and fingerprint similarity searches over loaded structures. Review the returned structure and descriptors such as molecular mass and calculated logP rather than assuming a match is conclusive.

Chemical similarity is an analytical aid, not a legal conclusion. Check whether the source actually associates the structure with a claim, example, compound, or other context relevant to the search question. Preserve the input structure and method used for each query.

Use the claim chart to map each claim element to evidence and record open gaps. Separate text that is explicit in the source from your interpretation. Capture source paragraph, page, or other locator so another reviewer can reproduce the reasoning.

The freedom-to-operate workbench helps organize claims for review, missing evidence, and human decisions. It does not infer legal status or issue a freedom-to-operate opinion. Export a source-grounded brief or investigation package that identifies the corpus, search method, sources, and limitations.

Marconi autosaves investigation changes and panel state, and it retains database-backed source information. If a conflict occurs, the application stops edits and offers local backup and recovery rather than silently overwriting another window. Keep exports and local database backups according to your organization's retention requirements.

External EPO OPS retrieval and optional AI access need configuration. AI analysis is deterministic by default and optional model requests require per-request consent. OCR, image-to-structure recognition, automatic legal-status interpretation, and autonomous FTO conclusions are outside the current implementation.

Create an investigation that states the invention, target, intended activity, jurisdictions, relevant dates, and question to be answered. A patent landscape depends on these boundaries: a different jurisdiction, family treatment, or cut-off date can produce a materially different set of records. Preserve the boundaries with the saved query and brief.

Use Marconi to organize a research screen and identify records for closer review. It is not a legal opinion service. Review the date and jurisdiction filters before interpreting an empty result or a supposed open area; a missing record can reflect source coverage or query scope rather than an absence of relevant rights.

Import source metadata through CSV, a portable investigation package through JSON, or private patent text/PDF through supported source ingestion. Check the import preview, citation fields, document identifier, jurisdiction, family, dates, and file status. Keep original source documents available for verification of parsed text and claim structure.

Public retrieval can use PubChem patent identifiers and EPO OPS when configured with authorized credentials and quotas. Retrieval is asynchronous and may succeed partially. Review the job status and provider provenance, then accept completed records into the investigation; failed or cancelled requests should remain visible rather than being treated as retrieved data.

The local search engine supports nested Boolean AND, OR, and NOT expressions, phrases, field restrictions, date filters, jurisdiction filters, and family deduplication. Use a sequence of narrow and broad queries when investigating terminology variants. Save the query and its source universe so a second reviewer can understand what the search actually covered.

Search ranking is explanatory lexical scoring. The interface may label semantic mode, but the implementation is a lexical approximation and does not include a bundled embedding model. A low-ranked record may still be important; inspect query matches and source text rather than treating ordering as relevance ground truth.

Open a record in the reader and move between bibliographic metadata, description, and claims. The claim explorer organizes dependencies and can identify explicit definitions, Markush alternatives, and numerical ranges. These are rule-based aids for locating structure in the text, not legal claim construction.

When a claim depends on another claim, follow the dependency chain and read the referenced language. Capture evidence against the exact source passage and distinguish a literal match, a possible technical relationship, and a missing disclosure. Every segmentation, element mapping, and interpretation should be reviewed by a qualified person before being used in an external legal analysis.

RDKit-backed tools can validate candidate SMILES or MOL structures and search loaded patent-associated structures using exact identity, SMARTS substructure patterns, a supplied scaffold, or fingerprint similarity. Inspect canonical structure and computed descriptors such as molecular mass and calculated logP where available. Document the comparison method and the corpus version used.

A fingerprint similarity score is a method-specific comparison of representations; it is not a legal determination that two compounds are equivalent or that one claim covers another. Structure search only operates over structures actually loaded into the investigation. Verify source attribution and chemical identity in the underlying record before citing a match in a claim chart.

Use family, citation, assignee, jurisdiction, and temporal views to organize the source corpus. Family deduplication controls how related filings appear in result lists; inspect the underlying publications when priority, national phase, or continuation details matter. Save landscape and white-space observations together with the query and source universe that produced them.

A cluster or empty region in a visualization depends on imported data coverage and classification. Marconi does not infer legal expiration, enforceability, ownership, or unclaimed subject matter unless those fields are explicitly present and verified. Treat visual patterns as prompts to inspect records and to refine the search, not as proof that a market or jurisdiction is clear.

The claim chart maps claim elements to invention context and evidence. For each element, record the supporting passage, related structure or technical object, missing evidence, and reviewer disposition. Keep “supported,” “not found in this corpus,” and “not assessed” distinct; absence from imported records is not proof that a disclosure does not exist.

The FTO workbench groups claims that require additional evidence and human judgment. Use it to prepare questions for counsel, capture the basis for an internal screen, and record which sources remain unchecked. It does not calculate infringement, invalidity, enforceability, or freedom to operate, and it should not be presented as a substitute for legal advice.

A research brief should state the invention question, search boundaries, queries, source set, family policy, notable records, claim-chart findings, chemistry method, and unresolved gaps. Inspect every citation against its source and include the date when a public record was retrieved. Keep interpretations clearly marked as researcher notes rather than provider facts.

Export CSV for metadata analysis, Markdown or JSON for structured handoff, print to PDF for a discussion document, or create a Qubosis evidence envelope for a downstream workflow. Exporting does not expand the source license. Check included private documents, source rights, and package contents before sharing outside the investigation team.

Investigation settings, query drafts, selected claim, reader section, document tabs, and panel settings autosave where supported. Source jobs retain explicit success, partial, failed, and cancelled states; completed content is added only after acceptance. This makes asynchronous retrieval and source review easier to distinguish.

If another window changes the same investigation, Marconi stops edits on the stale revision and offers a local backup and recovery path. Preserve unsaved local notes, reload the current state, compare the changes, and reconcile before continuing. The bundled training corpus contains fictional DEMO records; do not cite them as real prior art or external patent evidence.