KnowSynth: A Multi-Agent AI-Powered Smart Learning Assistant for Brazil's ENEM Exam

KnowSynth (also known as EduSynth) is a multi-agent AI learning assistant designed specifically for candidates preparing for Brazil's ENEM exam. Through collaboration among six AI agents, it can convert any exam topic into complete learning materials (including original questions, strategy analysis, and progressive problem-solving hints) in seconds. The project was developed by silasluiz96-alt, open-sourced on GitHub, and released on June 4, 2026.

ENEM is one of Brazil's most important university entrance exams, attracting millions of candidates each year. To address the huge demand for exam preparation, KnowSynth provides personalized intelligent learning support through generative AI. The original author/maintainer of the project is silasluiz96-alt, the source platform is GitHub, original link: https://github.com/silasluiz96-alt/KnowSynth, released on June 4, 2026.

The core of KnowSynth is a six-agent pipeline architecture:

1. Researcher Agent: Uses Tavily API for three-layer retrieval (teaching resources, latest news, academic references)
2. Critical Analysis Agent: Uses Groq's LLaMA 3.3 70B to evaluate exam relevance, common mistakes, etc.
3. Comprehensive Generation Agent: Generates structured learning materials (introduction, core points, mock questions, etc.)
4. Strategy Guidance Agent: Progressive hint system (three-level hints + final answer) to guide active thinking
5. Performance Analysis Agent: Tracks learning sessions and generates personalized reports
6. Orchestrator: Coordinates data flow between agents Tech stack: Python 3.12+, Streamlit, Groq API, Tavily API, etc.

Workflow after the user inputs a topic (e.g., "Fordism"):

1. The orchestrator starts the pipeline
2. The researcher agent performs three-layer data retrieval
3. The critical analysis agent evaluates exam relevance
4. The comprehensive generation agent creates learning materials
5. The strategy guidance agent provides progressive hints
6. The performance analysis agent generates learning suggestions Agent behaviors are defined via markdown files in the .claude/skills/ directory (e.g., researcher.md) and can be configured for iteration.

For Candidates:

• Personalized learning materials
• Progressive hints foster active thinking
• Understand exam key points and pitfalls
• Session reports optimize review For Educational Technology:
• Multi-agent division of labor improves quality
• Pipeline architecture enhances maintainability
• Progressive guidance balances answers and independent learning
• Configurable behaviors facilitate iteration.

Future Plans:

• User system: Google login, personal profile, session history
• Data analysis: Progress dashboard, performance charts
• Intelligent planning: Personalized schedule, active reminders
• Platform expansion: PWA, offline mode, PDF export, official question bank integration Open Source Community: Uses MIT license , PRs and Issue discussions are welcome, reflecting the trend of AI democratization (using open-source models and free API layers).

KnowSynth is an excellent case of multi-agent AI in educational scenarios, providing end-to-end support from information retrieval to strategy guidance. Design principles:

1. Decompose tasks to specialized agents
2. Clear data interfaces between agents
3. External files define behaviors for easy iteration
4. User-centric progressive hint design Similar applications in the future are expected to become important tools for personalized learning.