Using the Chatbot¶
Complete guide for interacting with the Poolula Platform AI chatbot to get answers about your business, properties, and documents.
Overview¶
The Poolula chatbot uses AI-powered search to answer questions by:
- Searching your database for properties, transactions, and financial data
- Searching your documents for formation docs, contracts, and policies
- Combining results to provide comprehensive answers with citations
Quick Start¶
Access the Chatbot¶
# 1. Start the API server
uv run uvicorn apps.api.main:app --reload --port 8082
# 2. Open browser
open http://localhost:8082
# 3. Ask a question in the chat interface
First Questions to Try¶
Property information:
- "What is our property address?"
- "What properties do we own?"
Financial data:
- "What was my rental income last month?"
- "Show me expenses for this year"
Documents:
- "What documents do we have?"
- "What's our business purpose in the operating agreement?"
How the Chatbot Works¶
graph TD
A[Your Question] --> B{Question Analysis}
B --> C{Choose Tools}
C --> D[Search Database]
C --> E[Search Documents]
C --> F[List Documents]
D --> G[Combine Results]
E --> G
F --> G
G --> H[Generate Answer]
H --> I[Show Response + Sources]
style D fill:#e3f2fd
style E fill:#fff3e0
style F fill:#f3e5f5
Behind the Scenes¶
1. Your question is analyzed
- AI determines what type of information you need
- Decides which tools to use (database, documents, or both)
2. Tools search for relevant information
- Database tool: Queries properties, transactions, obligations
- Document search: Finds relevant text in your ingested documents
- Document list: Shows available documents
3. AI generates an answer
- Combines information from multiple sources
- Provides citations showing where information came from
- Formats answer in clear, natural language
Types of Questions¶
Property Questions¶
What you can ask:
- Property addresses and details
- Acquisition information
- Depreciation basis
- Placed-in-service dates
Examples:
"What is our property's address?"
→ Searches database for property record
"What is the land basis for our property?"
→ Queries financial data from database
"When was our property placed in service?"
→ Retrieves depreciation start date
Financial Questions¶
What you can ask:
- Rental income and expenses
- Transaction history
- Category breakdowns
- Date-range queries
Examples:
"What was my rental income in August 2024?"
→ Filters transactions by category and date
"Show me all utility expenses this year"
→ Queries expenses by category
"What were my total expenses in 2024?"
→ Aggregates transaction amounts
Document Questions¶
What you can ask:
- What documents exist
- Content within documents
- Formation details
- Compliance requirements
Examples:
"What documents do we have?"
→ Lists all ingested documents
"What's our LLC's business purpose?"
→ Searches formation documents
"Who are the members of our LLC?"
→ Searches operating agreement
Hybrid Questions¶
What you can ask:
- Questions combining database + documents
- Cross-referencing multiple sources
Examples:
"What properties do we own and what documents mention them?"
→ Queries database for properties
→ Searches documents for property references
→ Combines both results
"Show me our EIN and where it's mentioned in our documents"
→ Gets EIN from database
→ Finds document references
Asking Effective Questions¶
Be Specific¶
Good examples:
- ✅ "What was my rental income in August 2024?"
- ✅ "Show me utility expenses for this year"
- ✅ "What is our property's land basis?"
Less effective:
- ❌ "Tell me about money"
- ❌ "What happened?"
- ❌ "Show me stuff"
Include Key Details¶
Time periods:
- "in August 2024"
- "for this year"
- "last month"
Categories:
- "rental income"
- "utility expenses"
- "property management fees"
Specific values:
- "our property's basis"
- "land basis"
- "building depreciation"
One Question at a Time¶
Good:
"What was my rental income in August 2024?"
... get answer ...
"What were my expenses in August 2024?"
Less effective:
"What was my rental income in August 2024 and what were my expenses and
what's my net income and what documents talk about rental properties?"
(AI can handle complex questions, but simpler is often better)
Understanding Responses¶
Response Format¶
Typical response structure:
[Answer text with relevant information]
Sources:
- [Source 1: Database query or document reference]
- [Source 2: Additional source if multiple tools used]
Source Citations¶
Database sources:
Means: Information came from querying the SQLite database
Document sources:
Means: Information found in specific document via semantic search
Interpreting Tool Usage¶
Look at the sources to understand how your question was answered:
Database query:
- Fast, precise answers
- For structured data (properties, transactions, dates, amounts)
- Returns exact values
Document search:
- Semantic search through text
- For unstructured information (formation details, agreements, policies)
- Returns relevant passages
Multiple sources:
- Hybrid query used multiple tools
- More comprehensive answer
- Cross-referenced information
Common Question Patterns¶
"What" Questions¶
Property info:
- "What is our property address?"
- "What is our EIN number?"
Financial data:
- "What was my rental income in [month]?"
- "What were my expenses for [category]?"
"Show" / "List" Questions¶
Transactions:
- "Show me all expenses in 2024"
- "List rental income by month"
Documents:
- "Show me what documents we have"
- "List all formation documents"
"When" Questions¶
Dates and deadlines:
- "When was our property placed in service?"
- "When is our annual report due?"
"How much" / "How many" Questions¶
Amounts and counts:
- "How much did we pay in utilities?"
- "How many transactions do we have?"
Tips for Best Results¶
1. Use Natural Language¶
You don't need special syntax - just ask naturally:
✅ "What's our property worth?" ✅ "How much rental income did we get last month?"
2. Start Broad, Then Narrow¶
Strategy:
1. "What documents do we have?"
→ See what's available
2. "What's in our operating agreement?"
→ Search specific document
3. "Who are the members listed in the operating agreement?"
→ Get specific detail
3. Check Sources¶
Always review the sources shown with answers:
- Verify information came from correct source
- Note relevance scores for document searches
- Cross-reference if needed
4. Rephrase if Needed¶
If an answer isn't helpful, try rephrasing:
First try: "Show me money stuff" → Too vague
Second try: "What was my rental income last month?" → Specific and clear
5. Use Context from Previous Questions¶
The chatbot maintains conversation context:
You: "What properties do we own?"
Bot: [Lists 900 S 9th St property]
You: "What's the land basis for that property?"
Bot: [Understands "that property" refers to 900 S 9th St]
Conversation Features¶
Session Context¶
The chatbot remembers previous questions in your conversation:
- References to "that property", "this document", etc. work
- Context helps with follow-up questions
- Maintained per browser session
New Conversation¶
Start a new session if:
- Changing topics completely
- Getting confused responses
- Want to clear context
(Refresh the page or use "New Conversation" if available)
Sample Questions by Persona¶
See Sample Questions for 133 example questions organized by:
1. New LLC Owner
- Formation and structure
- Annual compliance
- Basic operations
2. Bookkeeper
- Property basis & depreciation
- Transaction analysis
- Chart of accounts
3. Property Manager
- Property operations
- Vendor management
- Maintenance tracking
4. Tax Preparer
- Depreciation schedules
- Deduction tracking
- Basis calculations
- Tax form preparation
Limitations & Known Issues¶
Current Limitations¶
Time-based queries:
- "last month" works
- Specific months work: "August 2024"
- Relative dates might be less accurate
Calculations:
- Basic aggregations work (sum, count)
- Complex calculations may require follow-up
- Tax calculations not automated (yet)
Document search:
- Works on ingested documents only
- Scanned PDFs need OCR (not yet implemented)
- Images in documents not searchable
When Chatbot Can't Help¶
For:
- Creating new records → Use API or manual entry
- Updating data → Use API endpoints
- Deleting information → Use API with caution
- Complex multi-step workflows → Manual process
Remember: Chatbot is read-only - it searches and reports, but doesn't modify data.
Troubleshooting¶
No Answer or Empty Response¶
Possible causes:
- No data exists for that query
- Question too vague
- Tool selection error
Try:
- Rephrase question more specifically
- Verify data exists (check database or documents)
- Break complex question into simpler parts
Wrong Information¶
If answer seems incorrect:
- Check sources shown
- Verify data in source (database or document)
- Report issue if source is correct but answer is wrong
Slow Responses¶
Normal response time: 1-3 seconds
Slower responses happen when:
- First query after startup (model loading)
- Complex document searches
- Large date ranges
Chatbot Not Available¶
Check:
# Is server running?
curl http://localhost:8082/health
# Start if needed
uv run uvicorn apps.api.main:app --reload --port 8082
Advanced Usage¶
Asking About Specific Documents¶
"What's in the Articles of Organization?"
"Search the operating agreement for member information"
"Find insurance policy details in our documents"
Filtering by Date Range¶
Aggregating Data¶
"Show total expenses by category for 2024"
"Break down rental income by month"
"What's my average monthly income?"
Next Steps¶
Learn more:
- Managing Documents - Add more documents for richer answers
- Sample Questions - 133 example questions to try
- Evaluation - How we ensure chatbot quality
For developers:
- Evaluation Harness - Testing methodology
- API Reference - Direct API access
- Architecture - How it works under the hood