RIFFAI Atlas

RIFFAI Atlas is an AI-native policy intelligence engine that turns satellite data and socioeconomic signals into real-time simulations for navigating climate, infrastructure, and equity trade-offs.

Comment

Inspiration

We want to tackle the challenge proposed by Stanford Ecopreneurship, in which California is the 4th largest economy in the world—yet it is navigating a structural polycrisis: accelerating wildfires, grid instability, housing shortages, and widening inequality. The California Doughnut Data Report made this visible. It shows where we overshoot planetary boundaries and where communities fall short of basic needs.

But something critical was missing: infrastructure intelligence.

At the center of California’s transition is the energy grid. Renewable projects are waiting in multi-year interconnection queues. Transmission congestion is rising. Electrification demand from EVs and heat pumps is accelerating. Meanwhile, wildfire risk threatens grid reliability, and energy burden disproportionately affects vulnerable communities.

We were inspired by Nira Energy (YC W22), which tackles grid interconnection bottlenecks. But if Nira Energy were founded "today" in a world shaped by AI and planetary boundaries, what would it look like?

It wouldn’t just map interconnection points.

It would

  • Autonomously analyze grid constraints
  • Simulate policy interventions
  • Balance renewable expansion with wildfire exposure
  • Quantify tradeoffs between infrastructure investment and energy equity
  • Operate as a live intelligence layer for the Doughnut economy

That is the vision behind RIFFAI Atlas.

We wanted to build an AI-native system that solves California’s grid interconnection bottleneck while embedding it within Doughnut economics, ensuring that energy transition decisions are evaluated not just by megawatts, but by:

  • Climate resilience
  • Infrastructure stability
  • Equity in access
  • Social foundation impact

In short, we were inspired to reimagine energy grid intelligence as a policy simulation engine with an AI-powered operating system for navigating California’s transition within planetary boundaries.

How we built it

We built RIFFAI Atlas from scratch during TreeHacks using:

  • Synthetic county-level grid dataset (58 counties × 5 years)
  • OpenAI models for policy interpretation and explanation
  • Vercel for platform deployment

1. Data Modeling

We structured California’s grid transition around key energy-policy indicators:

  • Renewable share (%)
  • Grid load (GW)
  • Interconnection delay (months)
  • Transmission congestion (index)
  • Wildfire grid exposure (index)
  • Energy burden (% income)
  • Policy investment score

We normalized variables and built composite indices:

Grid Stability Score:

[ \text{Grid Stability} = R - (0.4C + 0.3W) ]

Where:

  • (R) = reserve margin proxy
  • (C) = transmission congestion
  • (W) = wildfire exposure

Clean Transition Score:

[ \text{Transition} = S + D - 0.2I ]

Where:

  • (S) = renewable share
  • (D) = distributed energy penetration
  • (I) = interconnection delay

Equity Energy Index:

[ \text{Equity} = E_t - 3B ]

Where:

  • (E_t) = equity transition score
  • (B) = energy burden

2. Simulation Engine (idea)

Policy sliders modify parameters through elasticities:

  • Renewable investment → increases renewable share
  • Grid investment → decreases congestion & delay
  • Fire mitigation → reduces wildfire exposure
  • DER incentives → increases distributed penetration

The simulation recalculates county-level outcomes instantly.

3. AI-Native Layer

We integrated OpenAI’s model to:

  1. Parse user natural-language policy requests
  2. Translate them into parameter adjustments
  3. Run the simulation
  4. Generate structured explanations of tradeoffs

This makes RIFFAI Atlas an AI-native system rather than a static tool.

Challenges we ran into

The biggest challenge we faced was time. Building a true policy simulation engine that meaningfully connects grid interconnection delays, wildfire exposure, renewable deployment, and equity indicators requires careful modeling. We initially aimed to implement a dynamic simulation layer with calibrated elasticities and multi-variable feedback loops, but designing a system that was both interpretable and realistic proved more complex than expected within a hackathon timeframe.

Another challenge was data structure. Even with synthetic datasets, organizing 58 counties across 5 years with grid-related indicators required thoughtful schema design to ensure the model could eventually scale into real CAISO and CEC data. We also had to balance ambition with feasibility—deciding what to prototype versus what to architect for later.

Finally, integrating the AI layer responsibly was non-trivial. We wanted the OpenAI model to reason over structured outputs rather than hallucinate policy conclusions. Designing guardrails around that interaction required iteration.

Accomplishments that we're proud of

Although we did not complete the full simulation engine, we successfully:

  • Designed a structured, energy-policy-focused dataset covering all 58 California counties across 5 years.
  • Built the architecture for an AI-native policy intelligence system.
  • Developed a working natural-language interface capable of interpreting grid and policy queries.
  • Mapped the structural relationships between renewable expansion, interconnection delays, wildfire exposure, and energy burden.
  • Framed grid interconnection not as an isolated engineering problem, but as part of California’s Doughnut economy.

Most importantly, we demonstrated a credible blueprint for what an AI-native grid and policy operating system could look like.

What we learned

We learned that modeling energy transition is fundamentally a systems problem. Grid interconnection, climate resilience, and equity cannot be separated—they form a coupled system. Simplifying those relationships into simulation-ready formulas requires more time and domain calibration than anticipated.

We also learned that building AI-native infrastructure tools requires constraint and discipline. Large language models are powerful, but they must sit on top of structured models to be trustworthy in policy contexts.

Finally, we learned that the real innovation was not in visualization—it was in reframing the grid interconnection bottleneck within planetary and social boundaries. Even without a finished simulation engine, we validated that this framing is both technically feasible and strategically powerful.

RIFFAI Atlas is not yet the finished operating system—but it is a strong architectural foundation for one.

What's next for RIFFAI Atlas

Post-TreeHacks, we plan to:

  • Replace synthetic datasets with live CAISO + CEC data.
  • Integrate wildfire satellite feeds from satellite providers in each region of operations.
  • Add time-series forecasting using LSTM models.
  • Develop an investment memo generator for policymakers.

Our long-term vision:

RIFFAI Atlas becomes the "global" AI-native operating system for navigating climate, grid, and equity tradeoffs.

California is the prototype.

The next version will reason over the planet.

Built With

  • openai
  • vercel
Share this project:

Updates