In Collaboration with Suman Kumar, Xiaodan Ren & Nischal Jakhar
E-Lense
E-Lense is a web-based, AI-powered, sustainable scheduling tool that monitors real-time electricity demand and carbon intensity, recommending optimal times to run large compute jobs at lower energy costs and with a reduced carbon footprint.
E-Lense was developed as the final capstone project for MIT IAP: Sustainable AI. The project focuses on demand shifting- optimizing computing schedules to better align with renewable energy availability, reducing the carbon impact of energy-intensive processes without placing additional strain on the grid.
Skills
Ideation, Project Management, Coding, Research
How might we shift AI compute to align with renewable energy availability and low grid demand?
User Inputs
Duration
GPU Information
Flex Window
Grid Operator
Job Priority
-
Minimize Job Delay
-
Prioritize Cost
-
Balance Cost and Carbon
The user can input details into E-Lense to optimize job prioritization, whether to minimize job delay, minimize cost, or both.
Backend Data Sources
-
US Energy Information Administration
-
Global Electricity Data Platform
E-lense uses electricity data provided from EIA and grid data provided from Electricity maps to generate optimized suggestions and recommendations.
The forecast is generated using a set of formulas we identified after comparing multiple carbon intensity models and evaluating their differences.
Price Model: Multipler = 0.7 - 1.5x (base on demand)
Example: Demand 20% ⬆ → 0.95 (Multiplier)
Base Rate: $0.13/kWh
Base Carbon Intensity: 385 gCO₂/kWh
Forecasts
-
Carbon Intensity
-
Price Based on Grid Demand
Output
The output provides data-driven evidence that quantifies carbon and cost emission reductions for a selected time period.
E-Lense enables users to make informed, data-driven scheduling decisions that:
-
Reduces emissions
-
Lower costs of training
-
Support ‘Sustainable AI’ initiative