On-device Visual Perception for Light-weight Browser Agents
Indian Space Research Organisation(ISRO)
Reasonable choice. The scoreboard liked it. The scoreboard is not the one asking questions on the day.
Worth considering. A genuinely interesting privacy-first architecture that browser APIs now make real, but reliable redaction is the whole game — one missed field breaks the premise, so make recall on sensitive elements your headline metric rather than showing it redacting only the obvious cases. Roughly 150–340 teams are expected to go here.
The receipts
Every red flag on this statement, in full. These are the four places it bites.
Exhibit A
The privacy premise demands near-perfect redaction, and one missed PII field defeats the entire point, so recall on sensitive elements is the make-or-break metric
It gets worse
Browser resource limits constrain how capable the local model can be, forcing a genuine accuracy-versus-footprint trade-off
Still reading?
Evaluation use cases are revealed only at the finale, so you cannot tune to the test and must generalise
And the finisher
Visual PII detection misses text rendered in images or canvas that DOM tags do not cover, which is exactly where redaction quietly fails
The damage report
Every score this statement earned, and what each one actually costs you.
Feasibility
3/5Buildable. Not comfortably. There is a week in here you have not planned for yet.
Transformers.js and ONNX Runtime Web genuinely run vision models in the browser via WebGPU, and DOM-based PII detection is tractable, but running a capable ViT client-side within browser resource limits while reliably catching every sensitive element is hard, and the redaction must be near-perfect because one missed PII field defeats the entire privacy premise.
Innovation scope
4/5There is something genuinely new here. Do not bury it under another dashboard.
The split architecture — local perception plus guaranteed client-side redaction before any server call — is a genuinely interesting and current design problem, and how you combine DOM tags with visual detection to catch sensitive elements reliably is open.
Clarity
4/5The ask is unambiguous, which quietly removes your favourite excuse.
The architecture, the client-side technologies, the redaction requirement and the local-versus-server split are described clearly, though the evaluation use cases are only revealed at the finale so the exact test is unknown.
Acceptance potential
3/5Middle of the pack. This statement will not win the room for you — you will have to.
The privacy-first framing is current and the in-browser architecture is genuinely interesting, but reliable redaction is the crux and one missed PII field breaks the whole premise, browser resource limits constrain the local model, and the finale-only evaluation means you cannot tune to the test — a demo that redacts the obvious fields but misses edge cases undercuts the core claim.
Effort
MassiveA semester of work wearing a hackathon costume. Something is getting cut; decide what now, not in week five.
In-browser model inference, reliable multi-modal PII detection and redaction, the local agent logic and the server-side coordination is a demanding full-stack build against real resource constraints.
Demo-ability
EasyEasy to demo — and so is everyone else's. Working is the floor here, not the achievement.
Watching sensitive fields get redacted in the browser before anything is transmitted, with the network payload shown to contain only sanitised data, is a concrete and convincing privacy demonstration.
The demo they will have already seen
Somewhere around 150–340 teams are heading here, and the description is doing the choosing for most of them. They will read the same brief, reach the same architecture, and build a version of the same demo you are planning. Being correct is the floor. If your five minutes could be swapped with the team before you and nobody in the room would notice, you have not picked badly — you have built predictably, which costs exactly the same and hurts more.
What survives
The ground worth standing on when the questions start.
- Transformers.js and ONNX Runtime Web genuinely enable in-browser model inference, so the client-side architecture is real rather than aspirational
- The redact-before-transmit demo is concrete and directly proves the privacy claim by showing the sanitised network payload
- DOM-tag awareness gives you structured, reliable signals for detecting sensitive fields alongside the visual model
Nothing here is fatal. It is just the list of places this statement pushes back, and you now get to push there first.
The framing is a joke. The findings are not — they are the same analysis on the statement page, and every line above is attached to a score or a fact in the record. It is one opinion with its reasoning attached, so argue with it before you trust it.