Compute Solver

Inspiration

Urban policy decisions—like transport pricing or fuel taxes—are often made without understanding how citizens will respond. We wanted to build a system that allows planners to simulate outcomes before implementing decisions, enabling data-driven governance instead of reactive policymaking.

What it does

Compute Solver is a City Digital Twin platform that allows planners to:

1. adjust prices for metro, bus, fuel, and water
2. simulate real-time demand and revenue impact
3. observe system-level effects like congestion and stress
4. identify optimal policies using an AI-driven optimization layer

How we built it

1. Generated a synthetic city dataset capturing population drift, congestion, and seasonal effects
2. Stored data using Delta Lake on Databricks
3. Trained a demand prediction model using Random Forest + MLflow
4. Built a simulation engine to model behavioral and system dynamics
5. Developed a Streamlit app for real-time policy interaction and visualization

Challenges we ran into

1. Handling Databricks Free Edition constraints (DBFS access issues)
2. Designing realistic synthetic data with meaningful interactions
3. Balancing model accuracy vs simulation interpretability
4. Ensuring real-time responsiveness within limited compute resources

Accomplishments that we're proud of

1. Built a fully working end-to-end system (data → ML → simulation → UI)
2. Integrated MLflow model lifecycle with a live application
3. Created a decision-oriented simulation, not just predictions
4. Delivered a real-time interactive planner experience

What we learned

1. The importance of combining ML with domain modeling (not just prediction)
2. How to build scalable pipelines using Databricks (Delta + MLflow)
3. Designing systems that are both technically sound and user-centric
4. Trade-offs between simulation realism and computational simplicity

What's next for Compute Solver

1. Integrate real smart city sensor data for live simulations
2. Add multi-objective optimization (revenue vs congestion vs sustainability)
3. Incorporate causal inference for policy impact estimation
4. Extend to multi-city and real-time streaming scenarios

Built With

• pyspark
• python
• sql