How avenue-growth-partners/agno-agentic-braylon works
Project Type and Purpose
• This is a modularized, AI-powered data-processing system (monolithic backend) for gathering and enriching intelligence on companies.
• It exposes a CLI and programmatic interface; it orchestrates multiple “AI agents” and integrates with external services (Grata, HubSpot, Jina, Serper, OpenAI).
File-Structure Analysis
• Top-level package (agp_intelligence) with all modules in one folder. No separate “frontend.”
• Key modules:
• main.py – CLI entry point / orchestrator
• config.py – environment-variable management
• utils.py – shared utilities (decorators, logging, error handling)
• integrations.py – API clients (Grata, HubSpot)
• tool_functions.py – low-level tools for agents
• agents.py – agent definitions & orchestration logic
• processor.py – batch/parallel processing engine
• agp_verticals.py – domain-specific enums
• Cross-cutting concerns: requirements.txt, logging (research_processing.log), output files, retry/rate-limit logic.
README Insights
• Describes six AI agents (Scraper, Software SaaS, AI-Native, Vertical SaaS, Enricher, HubSpot).
• Highlights features: parallel processing (batch size, workers), modular architecture, robust error handling.
• Emphasizes external integrations and final CSV/log outputs.
a. Main Components
1. CLI Layer (main.py)
2. Configuration Layer (config.py)
3. Processing Engine (processor.py)
4. Agent Orchestrator (agents.py + tool_functions.py)
5. External Integrations (integrations.py)
6. Utility Layer (utils.py, agp_verticals.py)
7. Data I/O (Input CSV ↔ Output CSVs, Logs)
8. External Services (OpenAI, Grata API, HubSpot API, Jina, Serper)
b. Relationships & Data Flow
• CLI Layer reads user params → passes to Configuration Layer
• Configuration Layer provides settings → invoked by Processing Engine
• Processing Engine reads Input CSV → divides into batches → spins up Worker Threads
• For each company in a batch: Processing Engine calls Agent Orchestrator
• Agent Orchestrator loads tool_functions & uses utils (retry, rate-limit)
• Agents invoke External Services via integrations.py
– Scraper Agent → Jina / Serper
– Classification Agents → OpenAI
– Enricher Agent → Grata API
– HubSpot Agent → HubSpot API
• Agent Orchestrator aggregates results → returns to Processing Engine
• Processing Engine writes progress files & final CSVs via Data I/O
• Logging events flow from all layers into research_processing.log (utils.py)
c. Architectural Patterns & Principles
• Pipeline / Batch-processing pattern
• Modular “agent” plugin pattern (separation of responsibilities per agent)
• Concurrency via worker pool (parallelism controlled by batch_size and num_workers)
• Resilience: retry, rate-limit, centralized logging
• Single Responsibility: each module handles one concern (config, utils, integration, etc.)
d. Technologies & Frameworks
• Python 3.x, pip dependencies (requirements.txt)
• OpenAI API (GPT-based classification)
• Grata API, HubSpot API, Jina API, Serper API
• CSV file I/O, built-in logging