Enterprise Knowledge Platform

Your wiki writes
itself. And gets
smarter every day.

The Enterprise Second Brain turns documents, decisions, and expert conversations into a living, compounding knowledge base — maintained by AI, governed by your team, trusted by your business.

team-feed · fortuity-prod
✦
Wiki page created — Progressive Carrier Appetite updated: liquor threshold revised 50%→35%
just now
⊕
Checklist answered — Restaurant BOP, 45% liquor, FL: 2 carriers eligible via IX API
1 min ago
◎
Pending review — New synthesis page: BOP Liquor Eligibility Matrix (3 pages cited)
3 min ago
⚑
Lint flag — carrier-appetite-liberty: stale claim, verify against March bulletin
8 min ago
847
Wiki pages
1,204
Queries today
12
Active tenants
The problem with RAG

Knowledge that compounds,
not knowledge that evaporates

Standard RAG retrieves from raw documents on every query — re-deriving the same connections every time. The Second Brain compiles knowledge once and keeps it current.

Traditional RAG
✕
Re-derives answers from scratch on every query
✕
Cross-document synthesis fails without perfect retrieval
✕
Nothing accumulates — same work repeated indefinitely
✕
No team visibility into what the AI knows or decided
✕
Knowledge quality degrades with document sprawl
Enterprise Second Brain
✓
Wiki compiled at ingest — synthesis already done
✓
Cross-references, contradictions, and connections pre-built
✓
Every query and document makes the knowledge base richer
✓
Team feed shows every decision, update, and open flag in real time
✓
Nightly lint keeps the wiki healthy as it grows
Platform features

Everything a knowledge base
needs to stay trusted

Six capabilities that separate a compounding knowledge platform from a document search tool.

⟳
Persistent wiki engine
The LLM writes and maintains wiki pages — entity pages, concept pages, decision logs, synthesis documents. Cross-references pre-built at ingest, not re-derived at query time.
⊞
Context engineering
Per-turn relevance re-scoring selects the highest-value wiki pages for each query. Token budget inspector shows exactly what's in context and why — visible in the Angular UI.
⚙
MCP tool chain
Connects to external APIs (carrier APIs, occupancy search, web) mid-query. Tool results are automatically staged as new wiki evidence — operational data compounds into knowledge.
◫
Team approval gate
All LLM-generated content enters pending status before going live. The team feed broadcasts changes via SignalR. Reviewers approve, reject, or edit before knowledge reaches users.
⬡
Multi-tenant isolation
Hard partition-key isolation in Cosmos DB and Azure AI Search. Each tenant gets their own domain schema, system prompt, and wiki page type definitions. Zero data leakage by design.
◑
Nightly lint engine
Azure Functions run a nightly health pass: contradiction detection, orphan pages, stale claims, missing cross-references. Flags surface in the team feed for human resolution.
How it works

Four operations.
One compounding asset.

The wiki grows through four automated operations — each one leaving the knowledge base richer than before.

1
Ingest — compile, don't just index
Drop a source document (PDF, carrier bulletin, claims export, meeting transcript). An Azure Function extracts text, runs the LLM synthesis pass, and creates or updates wiki pages — entity pages, concept pages, decision logs. A single source may touch 10–15 pages. Everything enters a pending state and waits for team approval before going live.
Azure FunctionsAzure AI Document Intelligence
2
Query — context re-evaluated every turn
Each user message triggers a hybrid BM25 + vector search, re-scores all wiki pages against the evolving conversation, and assembles the optimal prompt within the token budget. Pages that were relevant in turn 1 may be dropped by turn 5 as the conversation narrows — and new pages surface that weren't needed before.
Azure AI SearchContext engine
3
File back — answers become wiki pages
When the LLM produces a useful synthesis — a comparison table, a decision analysis, a new cross-carrier insight — it flags it with [NEW KNOWLEDGE]. The API automatically stages this as a pending wiki page. The team reviews, approves, and the answer becomes permanent institutional knowledge rather than disappearing into chat history.
Cosmos DBSignalR team feed
4
Lint — the wiki stays healthy
Every night, an Azure Function runs a full health pass across the tenant's wiki: it finds contradictions between pages, stale claims that newer sources have superseded, orphan pages with no inbound links, and important concepts that deserve their own page. Lint flags appear in the team feed for human resolution.
Azure Functions (timer)SignalR broadcast
Context engineering

Every token in context
earns its place

The context window inspector shows exactly which wiki pages were selected for a query, their relevance scores, and what changed from the previous turn.

Context window — Turn 3 87,420 / 128,000 tokens used
Token allocation
System
Wiki (45k)
History (23k)
Query (13k)
Output reserve
System prompt + schema Wiki pages Conversation history User query Output reserve
Wiki pages selected (10 of 847)
carriercarrier-appetite-progressive
0.91840 tok
carriercarrier-appetite-travelers
0.841,120 tok
rulerule-liquor-liability-limits
0.79590 tok
conceptliquor-liability-endorsements ↑ new this turn
0.77620 tok
entityoccupation-restaurant-bop
0.71760 tok
apiix-api-integration ↓ dropped (score 0.19)
0.19650 tok
query: "Which carriers write restaurant BOP above 45% liquor? Build a comparison table."
Insurance domain — underwriting checklist

Checklists generated from
your wiki, answered by tools

Submit a risk. The platform queries your wiki for relevant guidelines, generates a checklist, calls carrier APIs and occupancy tools to pre-answer items, and files the result back as institutional knowledge.

Restaurant BOP 45% liquor · 8,400 sqft · Tampa, FL
⚑ Yellow risk
✓
Is liquor revenue above the carrier threshold for standard BOP?
45% exceeds Progressive (35%) and Liberty Mutual (40%). Travelers (60%) and Hartford (50%) remain eligible. Two of four carriers ineligible.
carrier-appetite-progressive carrier-appetite-travelers ix_carrier_api
⚑
Is a separate liquor liability endorsement required?
Hartford requires a standalone liquor liability endorsement at this percentage. Travelers permits a sublimit within the BOP package. Underwriter should verify endorsement pricing with Hartford.
rule-liquor-liability-limits liquor-liability-endorsements
✓
Is the occupation class eligible for BOP in Florida?
GL Class 58121 (Restaurant, full service) is eligible in FL for BOP under both Travelers and Hartford. No state-specific exclusions on file.
occupation-restaurant-bop occupancy_search state-florida
◎
Any recent appetite changes affecting this class in FL?
Progressive tightened FL restaurant BOP threshold Q1 2026 (50%→35%). No changes on record for Travelers or Hartford in the last 90 days. Filed back as wiki page: decision-appetite-progressive-fl-2026-q1.
decision-appetite-progressive-fl-2026-q1 web_search
Multi-tenancy

One platform.
Every team isolated.

Each tenant gets their own domain schema, wiki conventions, system prompt, and MCP tool configuration. Cosmos DB partition keys enforce isolation at the data layer — not just the API.

Fortuity Underwriting
domain: insurance_underwriting
page types: carrier, occupation, rule, decision, checklist
mcp tools: ix_api, occupancy_search, web_search
wiki pages: 847
Legal Operations
domain: legal_knowledge
page types: statute, precedent, contract, jurisdiction
mcp tools: westlaw_api, web_search
wiki pages: 312
Competitive Intel
domain: competitive_intelligence
page types: company, product, market, signal
mcp tools: web_search, factset_api
wiki pages: 194
The compounding effect

The 500th query is smarter
than the first.

Every query that produces a useful answer, every document ingested, and every checklist completed makes the next query faster and more accurate.

Drag to see how the platform compounds over time
Queries answered on this tenant
50 queries
124
Wiki pages compiled
38%
Checklist items auto-answered
8.4
Avg relevance score
12s
Avg checklist generation time
Technical architecture

Azure-native.
Production-ready.

Five layers, fully managed on Azure. Python FastAPI orchestration, Angular 21 workspace, Cosmos DB for multi-tenant storage, Azure AI Search for hybrid retrieval.

UI
Angular 21+Signals + RxJSSignalR clientAzure Static Web Apps
API
Python FastAPIPydantic v2MSAL / Entra IDAzure Container Apps
Workers
Azure Functions v4Service Bus triggerAI Document IntelligenceTimer trigger (lint)
Storage
Cosmos DB NoSQLAzure AI SearchAzure Blob StorageRedis Cache
LLM
Claude Sonnet 4Azure OpenAI EmbeddingsGPT-4o-mini (lint)MCP tool chain

Ready to build a wiki that
thinks for itself?

Deploy the Enterprise Second Brain on your Azure tenant. Bring your documents. The wiki builds itself.