Official Judging System.
PROJECT: ECLIPSE uses a strict 100-point evaluation breakdown focused on autonomous reasoning, tool usage, system reliability, and working execution.
Evaluation Criteria & Weights
TOTAL: 100%Does the team clearly articulate the problem domain and why their solution addresses it?
Does the system demonstrate autonomous multi-step reasoning, decomposition and planning — not just prompt-response?
Does the agent call real tools, APIs or systems and act on the results?
Does the system complete meaningful end-to-end tasks without constant human prompting?
How well does the system handle unexpected inputs, API failures, ambiguous data and edge cases?
Does the solution introduce a novel approach, architecture or application of AI/agentic systems?
Is the demonstration clear, compelling and does it show the system working end-to-end?
What Judges Expect to See
INPUT
System receives a realistic real-world input
PLAN
Agent decomposes the task and selects tools
TOOL USE
APIs, databases or systems are queried
DECISION
Agent reasons over results and decides next action
ACTION
Agent executes or proposes a real-world action
VERIFICATION
Outcome is checked, failures handled, human alerted if needed
RELIABILITY MATTERS (15%)
AI agents operate in unpredictable environments. Submissions will be explicitly tested against edge cases, API rate limits, malformed inputs, and execution failures. Systems that gracefully handle failures and self-correct will earn top marks.