Arc|line

Renewable energy developers waste 6 weeks per site analysis. Arc|line's 8-agent AI pipeline cuts that to 11 seconds, scoring 60 sites instead of one.

Comment

Arc|line

Turning land into investor-ready solar projects in under a minute

Inspiration

Renewable energy doesn’t have a technology gap — it has a paperwork gap.

A single solar developer like Sarah might spend 6 weeks vetting one site:

  • pulling land records
  • calling utilities
  • modeling financials
  • drafting outreach

All before knowing if the project is even viable.

Most sites get dropped.
Most communities that could host clean power never hear about it.

We asked:

What if an AI could do all of that in 60 seconds — across 60 sites at once?

What It Does

Arc|line turns any land parcel into a complete, investor-ready solar project packet in under a minute.

A developer drops a pin on a map. Our 8-agent AI pipeline immediately:

  1. Scores the site
    Solar irradiance (GHI), slope, zoning, and land suitability
    → produces a 0–100 solar viability score

  2. Pulls land intelligence
    Owner name, parcel ID, acreage, last sale price

  3. Estimates grid costs
    Nearest substation, distance, and interconnection cost range

  4. Models the financials
    IRR, CAPEX, payback period, and IRA tax credit eligibility

  5. Drafts the outreach

    • Personalized landowner email
    • 12-page investor-grade PDF

Result:
A professional project packet — something that used to take a full analyst team six weeks — delivered in 11 seconds.

How We Built It

Architecture Overview

  • Backend: Python + Flask
    Orchestrates an 8-agent pipeline.
    Each agent:
    • calls a dedicated data source
    • emits structured JSON
    • hands off clean inputs downstream

Site Scoring

  • Google Earth Engine (Python API)
  • Data sources:
    • NASADEM → slope
    • NASA POWER → GHI
    • zoning + land-use overlays

Output: [ \text{Suitability Score} \in [0,100] ]

Land & Grid Intelligence

  • Parcel ownership via land-record APIs
  • Grid proximity via OpenStreetMap + Overpass
  • Interconnection cost estimated from: [ \text{Cost} \propto \text{distance to substation} + \text{voltage class} ]

Financial Modeling

  • Energy production via PV performance models
  • Regional electricity pricing
  • IRA Section 48 credit logic (including adders)

Key outputs:

  • CAPEX
  • IRR
  • Payback period

[ \text{IRR} = \arg\max_r \left( \sum_{t=0}^{T} \frac{C_t}{(1+r)^t} = 0 \right) ]

All numbers tuned to remain defensible within ±10%.

AI Generation

  • Natural-language site summaries
  • Structured PDF section generation
  • Personalized landowner outreach emails

Strict JSON schemas + output validation ensure:

  • consistency
  • layout stability
  • zero hallucinated fields

Frontend

  • React map interface for pin-drop workflows
  • Remotion for a fully animated pitch presentation
  • Flask API backing all real-time requests

Challenges We Ran Into

1. Heterogeneous Data Sources

Stitching parcel data, satellite imagery, grid maps, and financial inputs without latency blowups required:

  • async orchestration
  • aggressive caching
  • strict schema enforcement

2. Financial Model Accuracy

IRA adders (energy communities, low-income bonuses) depend on:

  • census tracts
  • year-specific rules
  • overlapping eligibility logic

Getting this right took significant iteration.

3. PDF Generation at Speed

LLMs can write prose — but investors expect:

  • charts
  • tables
  • consistent section structure

We built a custom templating layer to turn structured outputs into a polished 12-page packet.

4. Prompting for Consistency

Ensuring the AI always returned machine-readable JSON, not free prose, required:

  • careful system prompts
  • output validation
  • fallback retries

Accomplishments We’re Proud Of

  • 11-second end-to-end latency
  • 60-site batch analysis in a single session
  • Financial outputs that hold up to scrutiny
  • Landowner emails that read like they came from a seasoned developer
  • A complete 2-minute animated pitch built entirely in Remotion

What We Learned

Agent orchestration is an architecture problem, not just an AI problem.

Bad data at step 2 compounds into garbage at step 8.
Clean interfaces between agents mattered more than any single model choice.

We also learned this:

The PDF is the product.

Developers don’t want dashboards.
They want something they can email immediately — to a landowner or an investor — without touching it.

Designing around that artifact changed everything downstream.

What’s Next for Arc|line

  • Portfolio mode
    Run 500 sites overnight → wake up to a ranked top-20 with full packets

  • Live interconnection queue integration
    Flag sites near congested substations before modeling

  • Landowner response tracking
    Email → reply → LOI → lease execution in one platform

  • Utility-scale expansion
    Extend models from 5–50 MW to 100–500 MW projects

  • White-label API
    Let EPCs, financiers, and land funds run Arc|line on their own site lists

Arc|line exists to remove friction — not from solar technology, but from the human process around it.

Built With

Share this project:

Updates