Poolula Platform DocumentationΒΆ
Welcome to the Poolula Platform documentation. This system provides integrated property management, financial tracking, and compliance tools for Poolula LLC, a Colorado-based rental property business.
What is Poolula Platform?ΒΆ
Poolula Platform is a data hub and natural language query system that combines:
- Transaction Analysis - Automated categorization and querying of rental income, expenses, and capital transactions
- LLC Compliance Q&A - AI-powered assistant for answering questions about formation documents, operating agreements, insurance, leases, and tax obligations
- Verification System - Rigorous evaluation harness to validate AI responses against known correct answers
Key FeaturesΒΆ
π€ AI-Powered ChatbotΒΆ
Natural language interface for querying business data and documents:
- Ask questions like "What was my rental income in August 2025?"
- Search formation documents, operating agreements, and compliance records
- Get answers backed by database queries and document citations
- 4 persona-based help sections (LLC Owner, Bookkeeper, Property Manager, Compliance Officer)
π Transaction ManagementΒΆ
Comprehensive financial tracking:
- Import transactions from Airbnb, bank statements, and expense receipts
- Automated categorization with 30+ expense categories
- Accrual accounting support
- Full provenance tracking (data lineage for all transactions)
π Document ManagementΒΆ
Organized document storage and semantic search:
- Store and search LLC formation documents, insurance policies, leases, tax documents
- ChromaDB vector store for semantic document search
- Metadata tracking (doc_type, effective_date, version, confidentiality)
β Compliance TrackingΒΆ
Never miss a deadline:
- Track LLC compliance obligations (Colorado periodic report, tax deadlines, insurance renewal)
- Automated reminders
- Recurring obligation support (yearly, quarterly, monthly)
π Evaluation & Quality AssuranceΒΆ
Verify AI accuracy with dual specialized evaluators:
- General evaluator: 5 cross-domain business questions
- Airbnb evaluator: 15 rental income questions with CSV ground truth validation
- Multi-dimensional scoring (tool usage, content relevance, numerical accuracy within 1% tolerance)
- Multi-provider comparison (Anthropic, OpenAI, Ollama)
- Target: β₯90% overall accuracy
Quick LinksΒΆ
-
Getting Started
Install Poolula Platform and run your first query
-
Using the Chatbot
Learn how to ask questions and interpret AI responses
-
API Reference
Complete API documentation for all endpoints
-
FAQ
Common questions and troubleshooting
Architecture OverviewΒΆ
Poolula Platform uses a hybrid architecture:
graph TD
A[Data Sources<br/>Airbnb, Banks, Manual] --> B[Import Scripts]
B --> C[SQLite Database<br/>5 Tables]
C --> D[FastAPI Service]
D --> E[Chatbot RAG System]
D --> F[Frontend<br/>Vanilla JS]
D --> G[CLI Scripts]
E --> H[(ChromaDB<br/>Vector Store)]
style C fill:#4A7C59
style E fill:#2C5282
style H fill:#2C5282
Core Components:
- SQLite Database - Single source of truth for transactions, properties, documents, obligations
- FastAPI Service - REST API for all operations
- RAG System - Retrieval-Augmented Generation combining database queries + document search
- ChromaDB - Vector store for semantic document search
- Vanilla JS Frontend - Clean, framework-free web interface
Technology StackΒΆ
Backend:
- Python 3.13+ with
uvpackage manager - FastAPI (REST API)
- SQLModel (SQLAlchemy + Pydantic ORM)
- ChromaDB (vector embeddings)
- Anthropic Claude API (Sonnet 4.5 model)
Frontend:
- Vanilla JavaScript (no framework dependencies)
- HTML5 + CSS3
- Marked.js (markdown rendering)
Testing & Quality:
- pytest (β₯80% coverage target)
- Evaluation harness (β₯90% AI accuracy target)
Project StatusΒΆ
Current Phase: Phase 6-7 - DSPy/MLflow Integration
Current Phase Status
β Phase 0-2: Complete - Infrastructure, database, REST API (properties + chat), RAG chatbot with Anthropic Claude, evaluation harnesses, vanilla JS frontend
π§ Phase 6-7: In Progress - DSPy/MLflow Integration - Baseline RAG implementation with feature-flag wiring β - MLflow experiment tracking scaffolding β - True DSPy pipeline (retriever/reasoner/verifier modules) - Next - Cross-provider optimization and evaluation - Planned
π Phase 3-5: Future - Additional REST endpoints (transactions, documents, obligations CRUD), production hardening, advanced analytics
Completed:
β Database schema (5 core tables with full provenance tracking) β SQLModel models with migrations (Alembic) β FastAPI REST API (properties and chat endpoints operational) β Database query tool (SELECT-only, safe SQL generation) β RAG chatbot with Anthropic Claude (multi-provider support planned Phase 6-7) β Chatbot with conversation history and audit logging β Airbnb CSV import with accrual accounting and duplicate detection β Vanilla JavaScript frontend with 4 persona sections β Document ingestion with ChromaDB vector store β Dual evaluation harnesses: - General business evaluator (5 questions) - Airbnb income evaluator (15 questions with CSV ground truth) β MkDocs documentation site (46+ pages) β 15 utility scripts for data management and evaluation
In Progress (Phase 6-7):
DSPy Implementation Status
Current State: The DSPy integration scaffolding exists (apps/dspy/ directory) but the current implementation is a RAG wrapper baseline, not a true DSPy pipeline.
What Exists:
- β
DSPy dependency installed and configured (dspy-ai==2.5.0)
- β
Feature-flag wiring in API to toggle DSPy vs. baseline
- β
MLflow tracking scaffolding and artifact logging
- β
Evaluation harness for baseline vs. DSPy comparison
- β
Dataset manifest and snapshot utilities
What's Next (True DSPy Pipeline):
- Implement DSPy modules: Retriever β Reasoner β Verifier
- Replace RAG wrapper with actual DSPy Program classes
- Add retrieval/reasoning/verification as separate DSPy signatures
- Enable DSPy self-optimization and compilation
- Extend evaluation metrics beyond token hit-rate heuristics
Detailed Roadmap: See docs/planning/dspy-mlflow-plan-2025-12-09.md for complete implementation plan and architecture diagrams.
- Multi-provider LLM support (OpenAI, Ollama) - Planned Phase 6-7
- Cross-provider optimization and comparison
Next Steps:
- Complete remaining API endpoints (transactions, documents, obligations)
- Expand test coverage (β₯80% target)
- Production hardening (authentication, field validation)
Getting HelpΒΆ
- Installation Issues: See Installation Guide
- API Questions: See API Reference
- General Questions: See FAQ
- Bug Reports: GitHub Issues
Last Updated: 2026-01-07 Version: 0.2.0 Status: Active Development (Phase 6-7: DSPy/MLflow Integration)