Passline

Inspiration

There is no official certification or seal for ADA compliance in the United States. A business can open its doors and operate legally without ever being inspected. This is especially common for small businesses operating in older buildings, ones that predate the Americans with Disabilities Act entirely. And even when an owner wants to get compliant, professional ADA inspectors are scarce and expensive, pricing out the small businesses that need help the most. The cruel irony? That same business can still be sued. Any individual who feels disadvantaged by a building's infrastructure can pursue litigation, and the settlement often costs far more than the renovation would have. We built Passline to close that gap.

What it does

Passline is a web app that lets business owners conduct their own ADA compliance audit — no inspector required. The owner starts by filling out a short questionnaire about their property: building age, last renovation date, building type, state, number of floors, parking availability, and more. This information is fed into an LLM that generates a tailored rule set — the specific ADA standards, measurements, and thresholds that apply to this building. From there, the owner selects a module to audit — a discrete part of their property like an entrance, a restroom, a corridor, or a stairwell. They record a short video of that space directly in the app (or upload one). Passline processes the video by extracting frames, running them through a depth estimation model to measure key dimensions (door widths, clearance distances, grab bar heights, threshold rise, etc.), and feeding those measurements alongside the relevant ADA criteria into a second LLM. That model identifies any violations, cites the specific ADA standard being violated, and estimates a remediation cost. The owner can audit as many modules as they want. When they're done, they finalize the audit and receive a structured compliance report: a full list of violations, plain-language explanations, ADA citation references, and an itemized cost estimate for bringing the property into compliance.

How we built it

Frontend: React Native mobile app for video capture, questionnaire flow, and report display. For demo purposes, we currently run it as a web app.
Questionnaire / Rule Engine: A Gemini call takes building metadata and generates a structured JSON rule table of applicable ADA requirements and measurement thresholds.
Video Processing: A frame extraction pipeline filters out redundant and low-quality frames to keep only the most useful views.
Module Classification: A second Gemini pass takes a subset of the filtered frames and determines what type of module (restroom, hallway, stairwell, etc) the video is of.
Depth Estimation: Depth Anything V2 is applied to extracted frames to estimate real-world dimensions such as door widths, clearance zones, object heights, and spatial relationships.
Violation Detection: A final Gemini pass combines the measured dimensions, extracted features, and module classification with the applicable rule set to detect violations, generate ADA citations, and estimate remediation costs.
Report Generation: The system outputs a structured JSON audit report with per-module violations, citations, recommended fixes, and total cost rollups.
Backend: A Python/FastAPI server orchestrates the full pipeline, with MongoDB used for report and audit storage and Featherless API's OpenAI SDK used for narrative/report generation.

Challenges we ran into

Issues with Git deleted our entire UI, we had to restart from scratch
We had difficulty merging the UI and the backend, as we are unused to using React and Node.js
Because our project is so AI-heavy, we needed to find the best balance between retaining all information and saving credits. For example, we couldn't take every frame of each video and analyze them using Gemini, because it is simply too computationally expensive. We had to figure out which frames were worth keeping and how to condense the information we needed

Accomplishments that we're proud of

We're proud of the end-to-end pipeline working as a coherent product. A user walks into a room, records 30 seconds of video, and gets a structured violation report with ADA citations and cost estimates — that's a genuinely useful tool, not just a demo. We're also proud of the report format itself. We designed the output JSON schema carefully so that violations are non-overlapping, properly cited, and scoped to the specific module. The final report reads like something a professional inspector might produce — but it costs the business owner nothing beyond their time.

What we learned

We learned a lot about the gap between "AI can do this in theory" and "AI does this reliably in production." Getting consistent structured outputs, accurate spatial measurements from video, and correct ADA rule application all required much more engineering than the concept suggested. The compliance domain is also deeply specific — we came away with a real appreciation for how complex ADA standards actually are and how much ambiguity exists in applying them.

What's next for Passline

Parking lot module — one of the most commonly violated ADA areas for small businesses
Remediation contractor matching — connect business owners directly with local contractors for flagged violations
Compliance score over time — re-audit support so owners can track improvement
White-label reports — formatted PDF export suitable for sharing with landlords, attorneys, or insurers
Expanded state/local rule support — several states have ADA-plus regulations that go beyond federal standards
Security measures — Google Auth sign-in, data security, etc.
ADA-compliant website — ensuring Passline's own web presence meets WCAG 2.1 accessibility standards