Data Center Site Selection Multi-Agent Negotiator with AutoGen & Cerebras Inference
Utility grid operators, municipal planning boards, and data center developers frequently collide over power reservations. Build a multi-agent negotiation system using Microsoft AutoGen and ultra-fast Cerebras Inference to simulate multi-party land allocation and sub-station capacity bargaining. Produce balanced power purchase terms within 30 seconds.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an ultra-fast multi-agent negotiation platform for utility and site selection using AutoGen and Cerebras Inference.
How work is evaluated
Evaluates multi-agent negotiation convergence speed and agreement Pareto efficiency.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
Agreement Constraint Verification
Agreed MW must not exceed available grid MW
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Negotiation Convergence Speed
Number of dialogue turns required to reach stable consensus • target: 5 • range: 1-12
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Set up AutoGen conversable agents representing Developer, Utility, and City Council
Configure Cerebras Inference endpoints for high-throughput, low-latency multi-turn conversations
Implement formal constraint satisfaction protocols during negotiation loops
Extract consensus agreements with enforceable power allocation terms
Reference links and supporting material
Data center utility inter-connection negotiation scenarios and contractual power capacity constraints.
Challenge setup
Evaluates multi-agent negotiation convergence speed and agreement Pareto efficiency.
Challenge input
JSON with requested_mw, available_mw, max_rate_per_kwh
AutoGen
Provides robust multi-agent dialogue orchestration
Cerebras Inference
Delivers sub-second response latency for multi-turn negotiation
Evaluated output
JSON with agreement_reached (boolean), agreed_mw, agreed_rate_per_kwh, total_turns
- Agreed MW must not exceed available grid MW
- Number of dialogue turns required to reach stable consensus • target: 5 • range: 1-12
- Benchmark: MultiAgentBargainBench
- Negotiation Convergence Speed target: 5
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- AutoGen · Required
- Hume AI · Optional
- Cerebras Inference · Optional
Evaluation contract
- Agreement Constraint Verification · Weight 1
- Negotiation Convergence Speed · Weight 1
Metrics
- Negotiation Convergence Speed · Target 5
Benchmarks
- MultiAgentBargainBench
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with requested_mw, available_mw, max_rate_per_kwh
Output: JSON with agreement_reached (boolean), agreed_mw, agreed_rate_per_kwh, total_turns
Execution harness: Python
Evaluation tools
- AutoGen · Agent Framework
- Cerebras Inference · Inference Acceleration
Recipe freeze marker
No recipe freeze time is recorded.