AgentGrid is a multi-agent AI backend system that analyzes business data and generates actionable insights and strategies automatically.
It simulates a team of AI “employees” working together — from data processing to decision-making.
AgentGrid takes business data (JSON or CSV), processes it through a structured pipeline of agents, and returns:
- Analytics (revenue, cost, profit)
- Insights (problems + reasons, including trend-aware analysis)
- Strategies (pricing, growth, cost-cutting)
The system follows a multi-agent architecture:
-
Data Agent Ingests and normalizes input data
-
Analytics Agent Computes key business metrics (revenue, cost, profit)
-
Insight Agent (LLM) Analyzes current data along with historical trends to identify problems and reasons
-
Strategy Agent (LLM) Generates actionable business strategies based on insights
-
Orchestrator Coordinates the pipeline, applies guardrails, validation, logging, and integrates historical context
- Multi-agent AI system
- FastAPI backend
- Local LLM integration (via Ollama)
- JSON + CSV input support
- Input validation (Pydantic)
- Output validation
- Logging for pipeline tracking
- Guardrails for edge cases (e.g., zero revenue)
- Robust JSON parsing for LLM outputs
- Historical data storage and retrieval using PostgreSQL
- Trend-aware insights using previous analysis runs
- Hybrid architecture (deterministic calculations + AI reasoning)
- Backend: FastAPI
- Language: Python
- LLM Runtime: Ollama (llama3 / phi)
- Data Processing: Pandas
- Validation: Pydantic
- Database: PostgreSQL
- ORM: SQLAlchemy
- Multi-agent system design
- LLM integration in backend systems
- Hybrid architecture (deterministic + AI reasoning)
- Stateful AI systems using historical context
- Real-world AI reliability techniques (guardrails, validation, parsing)
- Separation of concerns (logic vs reasoning)