|
Turn ISO, IEC, IEEE, 3GPP, GSMA, RFC, Matter, Bluetooth or USB documents into queryable AI assistants engineers can actually rely on. Every answer grounded, every list complete, every requirement anchored to its section.
Standards work is not casual lookup. A single bad answer erodes trust in the entire system.
"List every shall in section 4."
Top-K retrieval returns 15 chunks. The section has 47 requirements.
"How many tables does Annex C contain?"
RAG cannot count what it did not retrieve.
"Which sections reference §6.3?"
RAG retrieves around the query, never back at it.
"Does this standard mention 5G NR anywhere?"
RAG happily invents a yes from adjacent text.
"Give me an example requirement for clause 7.2."
Composer LLMs invent canonical-looking requirements that never appear in the source.
Two pipelines, one product. RAG for depth, knowledge graph for coverage.
Drop in a PDF of an ISO, IEC, IEEE, 3GPP, GSMA, RFC, Matter, Bluetooth or USB standard. The curation pipeline runs once per document.
Page-aware extraction builds a structured graph of clauses, requirements (shall/must/should/may), definitions, tables and cross-references, keyed by the document's own section numbers.
Three optional Claude calls fix the three known regex blindspots: glossary tables, requirement validation, and table-title recovery. Cost: $0 to $0.80 per spec.
Every question runs RAG for depth plus a KG agent for coverage. The composer treats agent evidence as ground truth and cites every item, every time.
Completeness, counting, inverse and coverage questions are detected upfront and routed with explicit "enumerate fully" instructions. Web search and Deep Research are gated per portal.
"List every X" returns every X with section anchors, every time. No silent truncation, no missing items.
The trained assistant is system-prompted at training time to forbid inventing requirements, sections or citations.
Users see which tools the agent called, what it found, and which chunks the composer cited. Expandable for full transparency.
Every requirement and clause keeps its canonical §A.7, §B.3.8, §1.2.3 number. No sequential renumbering of front matter.
Inverse lookups, dependency chains and citation paths are first-class queries against the deterministic graph.
Each portal has its own branding, auth gate, standards selection, feature flags and guardrails. One platform, many tenants.
Each persona gets a purpose-built UI with the controls and visibility they need, and nothing else.
Engineers, conformance teams and standards authors using the branded chat surface. They see only the standards the admin enabled, with their org's logo and auth.
Configures one or more portals. Owns branding, authentication, standards selection, feature flags, guardrails and analytics for their tenant.
Curates the public standards library shared across every portal. The entire super-admin namespace is invisible to anyone outside the configured email domain.
SaaS, Private Cloud or AI-in-a-Box. Same product, same guarantees, your data never trains our models.
Hosted, multi-tenant, fastest to adopt
In your VPC, with your keys and IdP
Air-gapped for sovereign environments