Visualization
Interactive demo in progress
Agile Agents
Multi-agent framework achieving 25% on SWE-BENCH
PythonMulti-Agent SystemsSWE-AgentLLMsSoftware Engineering
[PLACEHOLDER] Built a multi-agent framework on top of SWE-Agent at UC Berkeley. Achieved 25% success rate on SWE-BENCH compared to 11% for single-agent approaches, demonstrating the power of agent collaboration for software engineering tasks.
SWE-BENCH Score
25% (vs 11%)
Approach
Multi-Agent Collaboration
Institution
UC Berkeley