Carbon-Aware Compute Orchestrator (CACO)

Multi-agent system for optimizing data center AI workloads with real-time grid energy signals
Submission for Digital Energy Grid Hackathon 2025 - Problem Statement 2: Compute-Energy Convergence

🎯 Problem Statement

AI and data centers are now major energy loads. How do we ensure our digital future doesn't destabilize the physical grid that powers it?

CACO transforms data centers from passive grid loads into active flexibility providers by intelligently scheduling AI workloads during low-carbon, low-cost periods while maintaining SLA compliance.

🌟 Key Features

🤖 Multi-Agent Architecture: Compute, Grid, and Coordination agents working in harmony
🌱 Carbon-Aware Scheduling: Minimize g CO₂/inference with real-time grid signals
💰 Cost Optimization: 30-40% reduction in electricity bills
⚡ Flexibility Monetization: Earn £150-250/MWh from P415 grid services
🔗 Beckn Protocol Integration: Publish compute flexibility as tradeable products
📊 Real-Time Forecasting: LSTM-based carbon intensity & price predictions
🔐 Blockchain Audit Trail: Immutable logs for P444 settlement compliance

🏗️ Architecture

┌─────────────────────────────────────────────────────────────┐
│                    COMPUTE AGENT                            │
│  Job Queue → Power Estimator → Flexibility Assessor        │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│                     GRID AGENT                              │
│  Carbon API → LSTM Forecaster → Opportunity Detector       │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│              OPTIMIZATION ENGINE                            │
│  Minimize: cost + carbon_penalty                            │
│  Subject to: deadlines, SLA, carbon caps                    │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│         COORDINATION AGENT (Beckn Protocol)                 │
│  Catalog Publisher → Order Handler → Settlement Verifier    │
└─────────────────────────────────────────────────────────────┘
                              ↓
                    Grid Operator (UKPN/ESO)

See full architecture diagram: docs/DESIGN_DOCUMENT.pdf

📁 Code Layout Highlights

src/domain/models.py - shared dataclasses (JobSpec, CarbonPoint, FlexOffer, etc.)
src/data_sources/ - async clients for Carbon Intensity API + Elexon BMRS.
src/compute_agent/, src/grid_agent/, src/coordination_agent/ - the three A2A agents.
src/optimization/engine.py - heuristic scheduler producing ScheduledJob + FlexOffer.
src/beckn/server.py - FastAPI Beckn BPP façade exported via python main.py beckn.
data/synthetic/workloads.json - sample AI workloads ingested by the compute agent.

🔌 Expected Agent Payloads

// Grid Agent
{ "command": "get_grid_forecast", "from": "...", "to": "...", "region": "GB" }

// Compute Agent
{ "command": "ingest_jobs", "jobs": [ ... JobSpec ... ] }
{ "command": "get_flexibility_profile", "from": "...", "to": "...", "cluster_id": "hpc-1" }

// Coordination Agent
{
  "command": "run_caco_planning",
  "horizon_hours": 24,
  "region": "GB",
  "cluster_id": "hpc-1",
  "optimization": { "carbon_penalty_weight": 0.5, "sla_penalty_weight": 2.0, "max_power_kw": 12000 }
}

Each agent responds with JSON strings; the coordination agent emits both optimized schedules and Beckn-ready flexibility offers which the Beckn server publishes as catalog items.

🔄 Beckn Protocol Flow (Spec-Compliant)

Every Beckn API (/search, /init, /confirm, …) immediately ACK/NACKs with {"message":{"ack":{"status":"ACK"}}} while the actual business payload is delivered asynchronously via the matching callback (/on_search, /on_init, /on_confirm).
context objects follow the Beckn Core 1.0.0 schema (domain, country, city, action, core_version, bap_id, bpp_id, transaction_id, message_id, timestamp, ttl). The transaction id stays constant across the entire flow; callback message IDs are regenerated.
/search requests expect Beckn’s canonical intent shape (message.intent.fulfillment.start.location/time, optional intent.tags for power/duration filters). The BPP turns the resulting flexibility offers into proper Beckn catalog items under message.catalog["bpp/providers"].
Example interaction:
1. BAP POST /search (intent describes time/location & tags like power_kw, duration_hours).
2. CACO BPP returns ACK immediately and, once the optimizer finishes, calls POST {bap_uri}/on_search with the catalog.
3. BAP POST /init with selected order; BPP ACKs and later invokes {bap_uri}/on_init.
4. BAP POST /confirm; BPP ACKs and follows up with {bap_uri}/on_confirm that includes the committed fulfillment.

This keeps the prototype faithful to the Beckn asynchronous contract while still allowing the coordination agent to perform multi-minute optimizations before publishing the catalog.

🚀 Installation

Prerequisites

Python 3.9+
Git
(Optional) Docker for containerized deployment

Clone Repository

git clone https://github.com/YOUR_USERNAME/deg-hackathon-carbon-aware-compute.git
cd deg-hackathon-carbon-aware-compute

Install Dependencies

# Create uv-managed virtual environment
uv venv
.\.venv\Scripts\activate  # macOS/Linux: source .venv/bin/activate

# Install project in editable mode
uv pip install -e .

Environment Setup

# Copy example config
cp config/config.example.yaml config/config.yaml

# Add API keys (optional for MVP)
# - Carbon Intensity API: https://carbonintensity.org.uk (no key needed)
# - Elexon BMRS: Register at https://bmreports.com

⚡ Quick Start

1. Generate Synthetic Data

python scripts/generate_synthetic_data.py --days 30 --output data/synthetic/

2. Launch the full CACO simulation

uv run python main.py launch --horizon-hours 24

3. Start individual agents (optional)

# in separate shells
uv run python main.py coordination
uv run python main.py compute
uv run python main.py grid
uv run python main.py beckn --port 8000

main.py is a Typer CLI, so uv run python main.py --help lists all available commands.

📚 Documentation

Design Document (PDF) - Complete hackathon submission
Architecture Guide - Detailed component breakdown
API Reference - Agent interfaces and Beckn endpoints
User Guide - How to use CACO in production
Contributing Guide - Development workflow

🌐 Data Sources

Source	Purpose	API/Link	Cost
UK Carbon Intensity API	Real-time & forecast grid carbon	https://carbonintensity.org.uk	Free
Elexon BMRS	Imbalance pricing, system frequency	https://api.bmreports.com	Free (registration)
Azure Workload Traces	Historical compute job patterns	https://github.com/Azure/AzurePublicDataset	Free
Google Cluster Traces	Job scheduling baselines	https://github.com/google/cluster-data	Free
Met Office DataPoint	Weather forecasts (cooling load)	https://www.metoffice.gov.uk/datapoint	£ (optional)

🗺️ Roadmap

✅ Phase 1: MVP (Weeks 1-2)

Synthetic data generation
Rule-based scheduler
Carbon Intensity API integration
Basic dashboard

🚧 Phase 2: ML Integration (Weeks 3-4)

LSTM forecasting model
XGBoost job predictor
CVXPY optimization solver
Performance benchmarking

📅 Phase 3: Beckn Protocol (Weeks 5-6)

Catalog publisher
Order handler (P415 activation)
Mock grid operator
Settlement verification

🎯 Phase 4: Pilot (Weeks 7-8)

University HPC cluster deployment
Real AI workload testing
Cost/carbon measurement
Blockchain audit trail

🚀 Phase 5: Production (Post-Hackathon)

Cloud provider partnership
Multi-region orchestration
Commercial SaaS platform
P415 market integration

🤝 Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

Development Setup

# Install dev dependencies
pip install -r requirements-dev.txt

# Run tests
pytest tests/

# Run linter
flake8 src/

# Format code
black src/

Current Needs

Data Scientists: Improve forecasting models
Energy Experts: Validate grid integration logic
DevOps: Kubernetes deployment, CI/CD pipeline
Frontend: Enhance dashboard visualizations

📄 License

This project is licensed under the MIT License - see LICENSE file for details.

TL;DR: You can use, modify, and distribute this software freely, even commercially, as long as you include the original license.

👥 Team

FlexCompute Labs - DEG Hackathon 2025

Name	Role	LinkedIn	GitHub
Xiaoyi Sun	AI/ML Engineer	https://www.linkedin.com/in/xiaoyi-sun-xs522	https://github.com/weepsdanky
Charles Cai	Energy Analyst	[Link]	https://github.com/charles-cai
Mathew Obasuyi	Backend Dev	https://gravatar.com/cyrilobasuyi1	https://github.com/Megabrain256
Khin Saw	Data Scientist	https://www.linkedin.com/in/khin-s-3946a161/	[link]

Contact: brainiac@hybridy.site | Discord: divineseal040

📖 References

Academic Research

Microsoft Research: Carbon Explorer (2024)
Meta AI: Chasing Carbon (2023)
Google: 24/7 Carbon-Free Energy Methodology

Standards & Protocols

Beckn Protocol v1.1
BSC P415 - VLP for DER Aggregation
BSC P444 - Half-Hourly Settlement

DEG Initiative

🏆 Hackathon Submission

Event: Digital Energy Grid Hackathon 2025
Problem: #2 - Compute-Energy Convergence
Platform: Dora Hacks
Deadline: 23 Nov 2025, 17:00 GMT

Submission Materials:

📊 Impact Metrics (Projected)

Metric	Value	Basis
Cost Reduction	30-40%	Shift to low-price periods
Carbon Reduction	60-70%	Schedule during renewable surplus
Flexibility Revenue	£200k/year	P415 market rates (2 MW capacity)
Grid Benefit	20-30 MW	Per 100 MW data center cluster
Emissions Avoided	150k tons CO₂/yr	If 10% of UK data centers adopt

🌍 Sustainability Statement

CACO directly contributes to:

UN SDG 7: Affordable and Clean Energy
UN SDG 9: Industry, Innovation, Infrastructure
UN SDG 13: Climate Action

By enabling 24/7 carbon-free computing and transforming data centers into grid stability assets, we're building the infrastructure for a sustainable AI future.

⭐ Star This Repo!

If you find this project interesting, please give it a star ⭐ and share it with your network!

Built with ❤️ for the Digital Energy Grid | Powered by Beckn Protocol | Carbon-Aware Computing FTW 🌱

Name		Name	Last commit message	Last commit date
Latest commit History 6 Commits
data/synthetic		data/synthetic
docs		docs
src		src
.env.example		.env.example
.gitattributes		.gitattributes
.gitignore		.gitignore
.python-version		.python-version
AI DEG Hackathon Design Document.pdf		AI DEG Hackathon Design Document.pdf
LICENSE		LICENSE
README.md		README.md
main.py		main.py
pyproject.toml		pyproject.toml
uv.lock		uv.lock

Uh oh!

Folders and files

Latest commit

History

Repository files navigation