-
Modular Multi-Agent Architecture: Distinct agents for hypothesis generation, external data integration, reflection, ranking, evolution, summarisation, semantic memory (using FAISS), and meta-review.
-
Real-Time Data Integration: Uses live APIs (Google Custom Search, NASA and arXiv) to fetch up-to-date research information.
-
Dynamic Adaptation & Self-Improvement: The Supervisor learns from meta-review feedback, automatically adjusting parameters (e.g., reflection token limits, score thresholds, and iteration counts) for improved performance over time.
-
Iterative Refinement Process: The system iteratively refines hypotheses using the Evolution Agent when initial outputs do not meet quality thresholds.
-
Parallel Processing & Caching: External data fetching and similarity retrieval are parallelised to reduce overall processing time. API responses are cached to avoid redundant calls, boosting efficiency.
-
Semantic Memory with FAISS: Incorporates an FAISS-powered Proximity Agent to retrieve similar past cases for enhanced contextual insight.
-
Comprehensive Summarization: Synthesises recent research findings with domain-specific summaries to provide a coherent final output.
-
Transparent Performance Metrics: Detailed logs and score evolution graphs provide clear insights into system performance and improvements over iterations.
-
User-Friendly, Interactive Interface: Built with Streamlit, the interface allows users to select topics, view detailed outputs, and explore a visual diagram of agent interactions.
