Document Intelligence Agent
Your Entire Policy Library, Answerable in Seconds
A conversational AI that reads your city's documents — zoning codes, ordinances, council packets, budgets — and answers questions in plain language with exact citations. Not a search bar. Not canned answers. It understands the content.
Document Agent
● Online
What's the minimum side setback for a detached ADU in an R-1 zone?
Understanding your question...
Searching zoning ordinance...
Synthesizing answer...
A detached ADU in an R-1 zone requires a minimum 4-foot side setback per Section 23C.24.040(B). This applies to all sides except where fire wall construction is used.
Sec. 23C.24.040(B) — Page 214
What It Reads
Every Document Your City Produces
Upload once — instantly searchable through conversation.
Council & Commission Packets
Agendas, staff reports, attachments from Granicus, Legistar, eSCRIBE
Ordinances & Municipal Code
Municode, AmLegal, or PDFs hosted on the city site
Resolutions & Policy Registers
Fee schedules, policy decisions, approvals from clerk's records
RFPs, RFQs & Contracts
Procurement/finance systems and shared drives including proposals received
General Plans & Zoning Maps
Comprehensive/specific plans plus land-use layers from planning's document center
Permits & Entitlement Records
Building/planning permits, inspections, conditions of approval
Budget, CIP & Financial Statements
Adopted budgets, CAFR/ACFR, capital project lists from finance repositories
Departmental Document Repositories
OneDrive/SharePoint folders for planning, public works, ED, city manager
Spreadsheets & CSV Data
Budgets, property lists, permit data, inventory — queryable via SQL
Live Web Pages
City websites, planning portals — fetched and indexed automatically
How It Works
From raw documents to instant answers
Every document type your city produces codes, packets, permits, budgets flows through a 5-stage AI retrieval pipeline and comes out as a cited, plain-language answer.
Ingests every document type your city produces
Semantic search + BM25 + reranking — not just keyword match
Returns plain-language answers and queryable tabular data

Why It's Better Than Search
5-Stage Retrieval Pipeline
Not a keyword match. A multi-model reasoning chain that finds the best possible answer from your documents.
1
Semantic Search
Finds chunks meaning what you asked, even if the exact words don't match
2
Keyword Search (BM25)
Catches exact terms and proper nouns — section numbers, ordinance names
3
Rank Fusion
Combines both result sets into one unified ranked list
4
Cross-Encoder Reranking
A second AI model scores each chunk: does this actually answer the question?
5
Context Expansion
Fetches the full surrounding section so the LLM has complete context to answer from
| Simple Keyword Search | Document Intelligence Agent |
|---|---|
| Matches exact words | Understands meaning and intent |
| Shows isolated snippets | Reads full sections for context |
| No continuity | Remembers conversation history |
| Can't handle tables or CSV | Queries tabular data via SQL |
| Returns many irrelevant results | Reranked to top-5 most relevant |
Key Benefits
What Makes It Different
5-Stage Retrieval Pipeline
Not a keyword search — semantic search, BM25, rank fusion, cross-encoder reranking, and context expansion work together to surface the most relevant answer every time.
Reads the Actual Ordinance
The full zoning code is indexed — not a simplified FAQ. When it cites 'Sec. 31-605(a)(1)' it pulled that text directly from your document.
Conversation Context
Ask a follow-up and it remembers what you asked before. No need to re-explain the situation with every message.
Handles Tables & CSV Data
Upload a property inventory spreadsheet and ask 'How many ADU permits were issued last year?' — the AI writes and runs a SQL query against it.
City-Specific, Always
Berkeley's rules aren't mixed with Riviera Beach's. Each city has its own indexed document set — answers are grounded in that city's actual documents.
Upload → Instantly Live
No re-training. No re-deployment. An admin uploads a new ordinance the day it passes and it's searchable within minutes.
Real-World Scenarios
Questions It Answers in Seconds
“How does the City sell surplus real estate?”
“What permits do I need to build an ADU?”
“Can I sell hot food on the street?”
“What's the minimum front setback in a residential zone?”
“What are the operating hours restrictions for outdoor dining?”
“What signage spacing is required for electric fences?”
“How many commercial parcels are over 10,000 sq ft?”
“What are the water conservation rules for landscaping?”
Who Benefits
One agent. Every audience.
| Audience | What They Get |
|---|---|
| City Planning Staff | Answer citizen and developer questions instantly — consistent, policy-correct, sourced. Reduce inbound emails on repetitive zoning questions. |
| Developers & Property Owners | Self-serve 24/7. Understand requirements before spending money on architects or permit applications. |
| Architects & Consultants | Cross-reference building codes and zoning rules for any city in one place, without digging through 300-page PDFs. |
| Citizens / General Public | Ask questions in plain English, get plain English answers. 24/7 access — not limited to counter hours. |
| City Leadership | Every answer is grounded in actual policy documents. Source citations mean answers are auditable and defensible. |
Impact
Consistent, cited answers — available the moment a new document is uploaded.
Planner Response Time
Days
→
Seconds
Inbound Repetitive Calls
High Volume
→
Self-Serve
New Ordinance Go-Live
Re-train & Deploy
→
Upload → Live
Give the Public 24/7 Access to Policy Answers
Grounded in your actual documents. Not AI guesswork. Every answer cited to the source.
