← all projects

gridwise

Operator notes in plain English become typed constraints, then an exact linear program finds the cheapest 24-hour energy plan.

  • python
  • fastapi
  • linear programming
  • llm

source ↗live ↗2026

Optimal 24-hour dispatch: grid import, solar, battery charge and discharge, with the operator's directive window highlighted
Optimal 24-hour dispatch: grid import, solar, battery charge and discharge, with the operator's directive window highlighted

What it is

Built for the BUP CSE Fest 2026 hackathon. A campus has solar panels, a battery and a grid connection, and an operator leaves notes like “keep at least 90 kWh in the battery from 6 PM to 10 PM”. GridWise turns those notes into a cheapest-possible plan for the day.

How it works

The rule is simple: the LLM never does arithmetic.

  1. An LLM (or a deterministic fallback when it’s slow or absent) classifies each note into one of six typed directives.
  2. Guardrails validate them: hour windows, factor bounds, percentage-to-kWh conversions.
  3. A HiGHS linear program with 96 variables and 73 constraints solves the exact optimum in under 5 ms.

Highlights

Three operator notes become two typed constraints and a no-op, and the API returns the plan totals
Three operator notes become two typed constraints and a no-op, and the API returns the plan totals

writeups about this