Mission-control dashboard — live negotiation HUD, Boston map, and acoustic telemetry
Round 4 — consensus corridor CLEARED (93/100): Jain fairness 1.0000, −15.9 dB noise reduction
Round 1 — direct baseline VETOED (14/100): the route cuts straight through community quiet zones
Skyline Consensus — three AI agents negotiate an eVTOL corridor Boston can actually accept

Skyline Consensus — multi-agent eVTOL flight-path negotiation for Boston

Inspiration

Boston is about to get a new sky. eVTOL air taxis and delivery drones are coming — but decisions about where they fly are being made behind closed doors. Logan is already one of the most noise-complained-about airports in the country, and the burden falls hardest on East Boston, Dorchester, Roxbury, and Mattapan. Community "input" today is a PDF comment form. We asked: what if AI could negotiate airspace the way a good city hall should — in the open, on the record, with residents actually at the table?

What it does

Skyline Consensus is a multi-agent negotiation platform for urban air mobility — and it runs live in your browser.

Three AI agents represent the real constituencies of Boston's low-altitude airspace — Community (noise equity and environmental-justice quiet zones), Operator (efficiency and revenue), and City (safety, compliance, and balance) — negotiating through a transparent, five-round protocol of structured proposals, objections, and vetoes. In the default scenario, the Round-1 direct corridor over the neighborhoods is vetoed at 14/100 consensus; by Round 4 the agents converge on a corridor that bends around the city: 93/100 consensus with a Jain's Fairness Index of 1.0000, sealed with a SHA-256 audit certificate. Residents click their neighborhoods to place quiet zones that become hard constraints the agents must resolve on the record. Every trade-off lands in an auditable ledger — a black box for airspace decisions.

How we built it

Mission-control dashboard in vanilla JavaScript + Leaflet + Canvas — zero frameworks, zero API keys, zero build step. Live Boston map with vertiports, EJ quiet-zone polygons, the vetoed direct path vs. the consensus corridor, a real-time telemetry HUD (altitude, speed, dBA), and an acoustic cross-section chart comparing noise before and after negotiation
Node.js/Express negotiation engine: three negotiators with explicit utility functions over noise, equity, efficiency, and compliance, driven by a transparent protocol (proposal → objection → weighted scoring → veto or consensus)
Acoustics + fairness math: distance-decay noise modeling with altitude attenuation, Jain's Fairness Index, and a Nash Bargaining solution scored before/after every round
SHA-256-signed audit certificate for the final package — every decision reproducible and on the record
CLI + verification suite: npm run simulate replays any scenario in the terminal; npm test verifies all five Boston scenarios
Boston geodata (vertiports, neighborhoods, EJ zones) modeled on public MassGIS, FAA, and Massport data
Built end-to-end with Cursor cloud agents — engine, dashboard, and docs shipped from one agent workshop

Challenges we ran into

Getting three self-interested agents to converge without collapsing into mush. We solved it with hard constraint floors and a veto rule: no package ships unless it beats the status-quo equity score. Modeling noise credibly in a weekend forced honesty about uncertainty — every number on screen shows its assumptions.

Accomplishments that we're proud of

A negotiation transcript you can actually audit — and an equity score that goes UP when the agents argue well. Round 1: 14/100 (Vetoed). Round 4: 93/100 (Cleared). Fairness: 1.0000.

What we learned

Fairness is a constraint-satisfaction problem before it's a sentiment problem — and auditable AI is more persuasive than confident AI.

What's next for Skyline Consensus

A pilot study with a city planning office or an AAM operator, and a standards push: the auditable trail doubles as a regulatory artifact for FAA AAM integration.