Atlas AI

Atlas AI is an enterprise decision intelligence platform built on top of the Mireye physical data API. Designed for high-stakes infrastructure siting ($100M+ Solar Farms, Wind Parks, Battery Storage, and Data Centers), Atlas AI bridges the gap between raw physical geospatial attributes and executive-level investment decisions.

Rather than overwhelming developers and investment committees with raw heatmaps or hundreds of uninterpreted GIS layers, Atlas AI runs pre-flight data quality auditing, calculates FERC-calibrated interconnection capex ($USD), computes net buildable parcel acreage, and retrieves RTO regulatory tariffs to emit clear, explainable site control recommendations.

Selecting sites for physical infrastructure is broken across three critical vectors:

• Silent Geocode Failure & Field Poisoning: Standard geocoders often snap vague inputs (e.g., city-name queries) to city centers or road dividers. Downstream spatial calculations (flood risk, terrain slope, transmission line distance) then run on random city-center points without alerting decision-makers.
• Euclidean vs Real Corridor Pathing: Physical data APIs return straight-line air distance to nearest transmission lines, completely ignoring environmental barriers (FEMA floodplains, USFWS wetlands, PAD-US protected lands) sitting directly in the interconnection path.
• RTO Tariff & Queue Congestion Friction: Project developers waste months submitting interconnection requests to substations locked in 36–54 month RTO queue backlogs (PJM, ERCOT, MISO) without early regulatory warning.

Atlas AI addresses these challenges with a multi-tiered pre-flight audit and decision harness:

• Centroid Misalignment Detector (centroidValidator.ts): Audits geocode match precision (rooftop vs. city-centroid) using 5-pattern pre-flight checks to prevent field poisoning.
• Grid Interconnection & Capex Engine (gridCapacityEngine.ts): Calculates corridor barrier cost multipliers (1.4×–1.9× for floodplains, wetlands, protected areas), voltage headroom, FERC-calibrated interconnection capex ($USD), and RTO queue risks.
• Physical Agent Buildable Area Mask Harness (buildableAreaHarness.ts): Deducts environmental and terrain hazard footprints to compute net buildable parcel acreage and emit actionable AI agent site control verdicts.
• Autonomous Land Acquisition LOI Engine (ownerOutreachEngine.ts): Converts spatial data and parcel owner lookup into formal, downloadable option-to-lease Letters of Intent (LOI).

Atlas AI uses a serverless Next.js 15 App Router architecture with strict server-side API proxies to protect API credentials and prevent client-side DevTools exposure:

User Request → Next.js 15 Workspace Route
                     │
        ┌────────────┴────────────┐
        ▼                         ▼
 Mireye Fetch API        Turso / LibSQL Edge DB
(300+ Raw Attributes)   (Vector Tariff Chunks)
        │                         │
        └────────────┬────────────┘
                     ▼
       Pre-Flight Audit Engine
 (Centroid Validator + Grid Capex)
                     │
                     ▼
       RTO Regulatory RAG Pipeline
(Gemini 3072-dim Vector Search + Groq Synthesis)
                     │
                     ▼
   Segmented Tabbed HUD & Executive Briefing

• Framework: Next.js 15 (App Router), React 19, TypeScript
• Database & Vector Storage: Turso (LibSQL edge database) storing 3072-dim embeddings
• AI & Vector Pipeline: Gemini gemini-embedding-001 for semantic vector search + Groq llama-3.3-70b-versatile for augmented regulatory briefings
• Testing & Reliability: Vitest suite with 62 automated unit tests covering spatial, financial, and centroid calculation engines

• Interactive Segmented Tabbed HUD: Allows decision-makers to toggle seamlessly between Overview, Centroid Precision, Buildable Footprint, Grid & Path, and Regulatory RAG Tariff views without cognitive overload.
• Live Regional Energy Market Tracker: Fetches live $/MWh Locational Marginal Pricing (LMP) across PJM, ERCOT, MISO, CAISO, and SPP to project annual solar generation revenues and data center power costs.
• Interactive Corridor Cable Path Visualizer: Renders an animated SVG transmission path from site origin through environmental barrier nodes to the grid substation with step-by-step cost callouts.
• Live Federal Map Layer & Reverse-Geocoding Inspector: Supports dynamic map tile switching (Vector OpenStreetMap, Satellite Imagery, USGS 3DEP Topography) and real-time Nominatim reverse-geocoding.

To surface authoritative regulatory guidance, Atlas AI embeds a curated knowledge base covering interconnection tariff rules, study deposits, and queue timelines across all 7 US RTO regions (ERCOT, PJM, MISO, WECC, SPP, NYISO, ISO-NE).

Queries are embedded using Gemini 3072-dim vector models, matched against Turso LibSQL cosine vector tables, and synthesized by Groq into site-specific regulatory briefs complete with source citations.

1. Preventing Geocode Match Ambiguity: When users query vague city names, standard geocoders return city center points with identical decimal precision. We engineered centroidValidator.ts to detect 5 misalignment patterns (road-snaps, missing ZIP/state, cluster outliers) and flag dangerous coordinates before field data is trusted.

2. Barrier-Aware Corridor Math: Calculating transmission line distance using straight-line Euclidean distance underestimates cost when wetlands or protected areas sit between the site and line. We implemented a barrier multiplier algorithm (1.4× for floodplains, 1.55× for wetlands, 1.9× for protected lands) to model FERC-calibrated interconnection capex ($USD).

All external API keys (MIREYE_API_TOKEN, GROQ_API_KEY, GEMINI_API_KEY, TURSO_DATABASE_URL) are strictly read on the server inside Next.js Route Handlers. No credentials use the NEXT_PUBLIC_ prefix, guaranteeing zero DevTools exposure or client bundle leaks.

Building Atlas AI reinforced that physical data becomes significantly more valuable when translated into explainable decisions rather than raw geospatial attributes.

Based on building Atlas AI's enterprise layer, we provided three core API feedback recommendations directly to Mireye's product team:

• Exposing geocode_match_type (rooftop, parcel_centroid, city_centroid) in /v1/geocode.
• Offering polygon-level field aggregates (e.g. % flood overlap, buildable acres) on /lookup.
• Incorporating environmental barrier awareness into transmission distance calculations.