Read slide text
Slide 1 of 29: NearLens: Smart glasses awareness
Bluetooth awareness for nearby smart glasses, with device-reported recording status where supported. Processed on the device.
By Marcos Rezende.
Available actions
- Watch the NearLens demo video.
- Visit Marcos Rezende's website.
Visual description
The NearLens icon and wordmark sit on the left of a dark background with a violet glow. On the right, a hand holds an iPhone displaying the radar, scan controls and violet detection cards. Small institutional logos appear at the lower right.
Slide 2 of 29: The challenge
How can AI assist in building a functional product that answers a real privacy problem?
Five phases of design thinking. AI at every step, a human decision at every gate.
The product
- NearLens. Awareness rather than surveillance, and nothing is uploaded automatically.
Stack
- React Native, Expo and Bluetooth Low Energy.
Distribution
- iOS App Store and Web, at nearlens.app.
Visual description
White text on a violet background. A Bluetooth symbol at the centre of concentric rings on the right illustrates listening for nearby radio signals.
Slide 3 of 29: Anyone nearby might be recording.
Nobody can tell. The system was designed that way.
The problem
- 1. Asymmetry: A phone shows it is filming. Glasses have a small LED that people nearby rarely notice.
- 2. Scale: Facial recognition from a single frame. Nairobi, 2024.
- 3. Gap: The only tool was Android and developer-only. Nothing for everyday iOS users.
Illustrated recording card
- Ray-Ban Meta. Recording, small LED.
- Captured, silently: Every frame stacks into a profile the person in it never agreed to.
- Consent asked: none. Visible signal: easy to miss.
Visual description
Three numbered points on the left face a stack of dark recording cards on the right. A white pair of camera glasses overlaps the cards. The stack represents observations accumulating over time, not a measured dataset.
Slide 4 of 29: The sentence that guided every decision
People in public spaces need a way to know when smart glasses are present, because silent capture creates an asymmetry that makes any informed decision about consent impossible.
How might we
- How do we inform without enabling harassment of wearers?
- How do we make detection usable for people with low tech literacy, including older adults?
- How do we set honest expectations when detection is probabilistic?
- How does a privacy app earn trust while asking for Bluetooth access?
Visual description
The problem statement is highlighted in a broad violet banner. Four question cards sit below it in two columns on a dark background.
Slide 5 of 29: How it actually went
Process: 30 days, two phases.
Design thinking was the scaffolding. The phases overlapped and doubled back.
Timeline, in row order
- Talking to owners: 7 interviews, with Claude.
- Writing the problem: 1 statement, with Claude.
- Killing ideas: 3 directions cut, with ChatGPT.
- Shipping: 328 commits, with Replit Agent.
- Testing in public: 6 observations across 4 cities.
- Rebuilding on Signals: 82 commits.
What caused the process to double back
- P3 would switch off Bluetooth if they knew. That sent us back to the problem statement.
- 823 devices and zero glasses at Meridian made us rewrite what the app claims.
- Naming took two rounds. AI proposed puns, I rejected every one.
Visual description
Horizontal bars are staggered across Phase 1 and Phase 2 rather than forming a strictly sequential staircase. Shipping has the darkest and longest bar. Three notes run along a dark footer.
Slide 6 of 29: Who we interviewed
Empathy: interviewee profiles. 7 interviews, by video call, approximately 30 minutes each.
Participants, quotes and reaction labels
- P1: Waterloo, Canada; former Meta staff. “The LED can be disabled in software.” Reaction label: Positive.
- P2: Atlanta, USA; Ray-Ban Meta Gen2. “I never read the T&Cs. I don't think about people when I film.” Reaction label: Neutral.
- P3: San Francisco, USA; UX designer, 2 years. “If I knew I could be detected, I'd turn off Bluetooth.” Reaction label: Negative.
- P4: Indianapolis, USA; approximately 85% while driving. “I avoid the technical details to keep my peace of mind.” Reaction label: Negative.
- P5: Ottawa, Canada; former photographer. “I'm recording without them knowing. With a phone, they know.” Reaction label: Neutral.
- P6: Ottawa, Canada; tech professional, wears them all day. “Detecting the glasses. Not accessing them.” Reaction label: Positive.
- P7: Toronto, Canada; owner for 3 to 4 years. “It's their device, I have mine. It's a fair game.” Reaction label: Positive.
- Conclusion printed below the profiles: Three fair, two against, two conditional. No majority, so NearLens informs rather than enforces.
Visual description
Seven participant rows occupy the left, each with an anonymised P-number, quote and coloured reaction badge. A large blue-violet portrait of a person wearing glasses fills the right.
Slide 7 of 29: Four insights that shaped the product
Empathy: core findings.
Four findings
- 01. The gesture: Invisibility is why they buy. Owners name it as a reason to buy. Making capture visible removes what sells the product.
- 02. The terms: None of the seven read them. No habit, fear of what they would find, or the sense that reading means quitting. One skimmed and kept only “anonymous model training”.
- 03. The reaction: Detection splits the room. Three fair, two against, two conditional. Acceptance comes from device symmetry, not a right to consent.
- 04. The LED: Beatable, but not trivially. Modified firmware, a cover, or plain daylight. Black tape disabled the device outright.
Visual description
Four white folder-shaped cards are arranged left to right on a light background, with violet numbering and labels. Pale concentric circles are decorative.
Slide 8 of 29: Six patterns across seven owners
Research insights.
Patterns
- Glasses hide the filming: A phone is visible when it films. Glasses never are.
- Nobody reads the terms: None of the seven, for three different reasons.
- Months of frames add up: The worry is the profile built up over time.
- The mic is the forgotten vector: Everyone fears the camera. The always-on audio worries whoever knows the hardware.
- Nobody has ever asked them to stop: Not one request across seven interviews. Discomfort shows as a look.
- Place decides what is fine: Off in bedrooms, childcare, clinics. On while driving and outdoors.
Visual description
Six white cards form two rows of three on a light background. Each has a violet outline icon, a bold finding and a short explanation.
Slide 9 of 29: Seven people. Four numbers that moved the product.
Empathy: what it changed.
Numbers, interpretation and product decisions
- 0/7 read the terms in full. Not one of the seven owners. So onboarding is one tap, with no jargon on screen.
- 3 · 2 · 2: Fair, against, conditional. Reactions to being detected. No majority either way. So the tone is awareness, never enforcement.
- 0 times a wearer was asked to stop. Discomfort shows as a look, never a request. So the app informs you. It never asks you to confront anyone.
- 0 servers in the design. The fear is the profile built over months, not one photo. So the log lives and dies on the phone.
Visual description
Four large lavender numbers head four columns on a dark background. A stylised rising line-chart object and radio-wave arcs decorate the upper right; they do not add quantitative data.
Slide 10 of 29: Three ideas were cut before the app
Ideation.
Discarded directions
- Browser extension: Wrong data. It found cameras, never glasses.
- Sightings map: Needed an account. A privacy app cannot ask for that.
- BLE dongle: Buildable, but too much for one month alone.
The decision
- NearLens: iOS, passive Bluetooth, runs on the user's phone.
- Uses the Bluetooth already in every iPhone. No server, no collection, free.
Visual description
A triangular violet beam narrows from the discarded directions on the left toward a point labelled “Converged on one app” on the right. A decision card spans the bottom.
Slide 11 of 29: How NearLens works
Passive Bluetooth scanning. Manufacturer matching. Zero data collected.
Flow
- In the room: Glasses broadcast anyway. Many Bluetooth devices broadcast a maker ID or name, whether NearLens exists or not.
- On the user's phone, Listen: Tap Scan. iOS asks for Bluetooth access. No pairing.
- Match: Checked against the known makers: Meta, Snap, Samsung, Google.
- Alert: A vibration and a notification, with the log written locally.
- Scans while the app is open. No server. Nothing uploaded automatically.
Visual description
Arrows connect an unboxed broadcast explanation to three violet cards labelled Listen, Match and Alert. A bracket above those three cards identifies them as running on the user's phone.
Slide 12 of 29: A match becomes one notification
The alert. Manufacturer, signal, one way out. Nothing else.
Example notification
- Smart glasses detected. Apple Inc., −68 dBm. A close control and a Snooze action are visible.
Design choices
- Shows why it matched: A maker ID or a name pattern from the device's own broadcast.
- Haptic first, screen second: A short in-app banner, with every match kept in the log.
- Snooze, not silence: One tap to stand down without turning detection off.
Visual description
A white notification card is shown on the left over violet concentric rings. A woman looking at her phone fills the right. The image illustrates receiving an alert; it is not evidence of a particular detection.
Slide 13 of 29: When a device declares it is recording
Version 2.4.2: PMSG.
Some glasses can send a Bluetooth signal that says “I am recording”. Since version 2.4.2, October 2026, NearLens reads it.
Meaning and limits
- What PMSG is: A proposed format by Paul Stefaan Mooij. Not an official standard yet.
- What the user sees: The signal, shown as the device's own claim.
- The limit: Glasses that do not send the signal are not caught this way.
Example card
- Simulated PMSG device. Simulation (0x0000). Signal −60 dBm. Estimated proximity approximately 3 to 7 metres.
- Privacy signal: Recording.
- This device reports its camera is recording. Reported by the device in its Bluetooth advertisement.
Visual description
A hand holds a white keychain with a red no-photography symbol. The simulated device card floats above the keychain. The illustration does not show that the keychain blocks recording.
Slide 14 of 29: First contact, zero friction
Screen 01: Welcome. Clear purpose before the first scan.
Design annotations
- 1. Headline: “Smart glasses awareness” sets the ceiling on what the app claims.
- 2. Single CTA: “Start scanning” centred in the thumb zone. Four tabs underneath it.
- 3. Onboarding: Five steps explain the app and ask for Bluetooth.
- 4. Zero login: No account, no email. The app opens ready to scan.
Phone screenshot
- NearLens. Smart glasses awareness.
- Turn on Bluetooth on your phone before starting.
- Keep your phone close to the area to scan.
- Walk around to detect nearby devices.
- You'll be alerted when smart glasses are nearby.
- The visible primary button reads “Scan Devices”. Bottom tabs: Scan, Intelligence, Logs and Settings; Scan is selected.
Visual description
An iPhone mockup sits on the left. Its dark welcome screen has a glowing blue-dot pattern, the NearLens icon, four guidance rows, a violet primary button and a light bottom tab bar. Four numbered explanations are on the right.
Slide 15 of 29: No match, stated plainly
Screen 09: Scan, nothing found. Scanning, five devices in range, no glasses.
Design annotations
- 1. Sweep, not a hunt: The radar reads as listening: slow, one colour, no targeting language.
- 2. Three honest counters: Devices, detections, scans. Zero means no recognised match this session, not that no glasses are near.
- 3. Pause and Stop: Stop is rose because it ends the session. Pause is neutral because it does not.
- 4. Reduced motion: The sweep stops when the iOS accessibility setting is on.
Phone screenshot
- NearLens. Scanning for smart glasses. Pause and Stop.
- Counters: 5 Devices, 0 Detected, 14 Scans.
- Empty state: Scanning for smart glasses. 5 devices in range, no glasses detected.
- Bottom tabs: Scan, Intelligence, Logs, Settings; Scan selected.
Visual description
A violet radar fills the upper part of the phone. A white content area below contains the counters and a quiet empty state rather than a warning. The iPhone sits left of four numbered annotations.
Slide 16 of 29: The card shows why it matched
Screen 10: Scan, detected. Two devices matched, both explained on the card.
Design annotations
- 1. Violet card, no siren: Matches slide in under the counters in brand violet. No modal, no interruption.
- 2. Why it matched: The card names the manufacturer and the rule that matched it, so the user can disagree.
- 3. Signal, not location: Signal and rough distance. Never direction or position.
- 4. Unknown stays unknown: An unrecognised device is labelled unknown instead of being guessed at.
Phone screenshot
- NearLens. Scanning for smart glasses. Pause and Stop.
- Counters: 5 Devices, 2 Detected, 12 Scans. Detected Devices: 2.
- First card: Ray-Ban Stories, Meta Platforms Inc. Manufacturer ID matches Meta Platforms Inc.; device name matches “ray-ban”. Signal −67 dBm, estimated proximity approximately 3 to 7 metres.
- Second card: Unknown device. Its lower details are partially obscured by the bottom navigation in this screenshot.
- Bottom tabs: Scan, Intelligence, Logs, Settings; Scan selected.
Visual description
The phone keeps the same radar and counter layout as the no-match state. Two violet cards now occupy the content area. Signal-strength bars are shown, not map pins or directional arrows.
Slide 17 of 29: A timeline the user owns
Screen 11: Logs. Every detection, with export and a clear-all.
Design annotations
- 1. 24h / 7d: One chart, two ranges. Enough to spot a pattern, not a dashboard.
- 2. Two numbers only: Entries and unique devices. Six from six devices is not six from one.
- 3. Chronological, colour-coded: Time, device, manufacturer, signal, estimated distance. One colour per manufacturer.
- 4. Export or wipe: Share the log as text, or delete it. Nothing is uploaded automatically.
Phone screenshot
- Detection Log header with share/export and trash controls.
- Detections chart with 24h and 7d choices; 7d is selected. A single violet bar appears at Today; the screenshot does not print a numerical vertical axis.
- 6 Total Entries. 6 Unique Devices.
- Records under WED · AUG 5, each at 7:09 AM: Spectacles 3, Snap Inc., −69 dBm, approximately 3 to 7 metres; Apple Glasses, Apple Inc., −42 dBm, approximately 1 to 3 metres; Ray-Ban Stories, Meta Platforms Inc., −67 dBm, approximately 3 to 7 metres; Unknown device, Unknown, −75 dBm, approximately 7 to 15 metres.
- A further Unknown device attributed to Apple Inc. is partly visible near the bottom; the screenshot does not expose all six records.
- Bottom tabs: Scan, Intelligence, Logs, Settings; Logs selected.
Visual description
The iPhone has a violet header and white timeline. Coloured dots beside records distinguish manufacturers. The slide's explanations are on the right of the phone.
Slide 18 of 29: Patterns, computed on the phone
Screen 12: Intelligence. A seven-day read of the user's own log.
Design annotations
- 1. Activity by hour: The week read hour by hour, with the peak named in plain text.
- 2. Four facts, four cards: This week, today, busiest day, peak hour. One number each, nothing to interpret.
- 3. On-device only: Apple's on-device models where supported, plain statistics elsewhere. No cloud.
- 4. Feeds Ask NearLens: Natural-language Q&A through Apple Foundation Models, on supported devices with Apple Intelligence available.
Current phone screenshot
- Privacy Intelligence.
- Most active day: Friday, 2 detections.
- Peak hour: 8 AM to 9 AM, 5 encounters.
- Top brand this week: Sample Brand Alpha, 5 detections.
- The question field contains “How close were the glasses?” beside a violet upward-arrow send button.
- Disclaimer: “On-device only. Answers may not always be accurate.”
- The iOS keyboard is open. No generated answer is visible in this slide.
Visual description
The updated slide shows an iPhone on the left with three visible summary cards, the question composer and keyboard. The four feature explanations on the right describe the broader screen, including summaries not currently in view.
Slide 19 of 29: Control without complexity
Screen 13: Settings. Every switch is local, and says what it costs.
Design annotations
- 1. Save detection history: Kept on the phone for up to 90 days or 500 entries. Turn it off and pattern alerts go with it.
- 2. Pattern alerts: Repeat encounters and dense clusters, with “runs on-device” stated where doubt arises.
- 3. Scan interval: 30 seconds by default, in 10-second steps. The user picks detection rate versus battery.
- 4. Known manufacturers: The list the app matches against, printed in the open.
Phone screenshot
- Settings. Privacy: Save Detection History and Pattern Alerts are switched on.
- Save Detection History: Store detections locally for up to 90 days. Required for pattern detection. Disable to keep history in-session only.
- Pattern Alerts: Get alerted when repeat encounters, dense clusters, or anomalies are detected. All analysis runs on-device.
- Scan interval: Scan every 30 seconds, with minus and plus buttons.
- Known manufacturers: Meta Platforms Inc., Ray-Ban Meta Smart Glasses, 0x01AB; Apple Inc., Apple Glasses, 0x004C.
- Bottom tabs: Scan, Intelligence, Logs, Settings; Settings selected.
Visual description
A violet settings header tops a white screen. Two switches, a numeric stepper and a manufacturer list make the controls visible. Explanations sit to the right of the iPhone.
Slide 20 of 29: The system came out of the screens
Signals, Version 1.1. W3C Design Tokens format.
Three steps
- 01. Ship first: Navy chrome, rose alerts, pill buttons: decisions that survived a real screen.
- 02. Audit our own code: Every hardcoded value named once, in one token file. 60+ web parts, 9 native.
- 03. Hand it to the build: The build fails on a stray hex or stale theme, so the audit holds.
- The screenshot is labelled “Signals: the audit, navigable” and links to the Signals Design System in a new tab.
Visual description
The Signals logo, version and W3C-format badge appear above three numbered steps. A browser screenshot on the right shows the Signals Design System overview, a dark navigation sidebar and component cards. Readable headings include “Signals Design System”, “A privacy app needs quiet certainty” and “Users open this app because they feel watched.” Smaller component-preview copy is not legible at this slide's scale.
Slide 21 of 29: Four tensions, four decisions
Ethics.
Tensions and decisions
- Tension 1. Alert without alarming: Rose instead of red, vibration rather than a siren.
- Tension 2. Detect without surveilling: Zero analytics, telemetry or server calls.
- Tension 3. Accessible, never technical: One tap. Accessibility considered throughout the design and implementation.
- Tension 4. Warn without inciting: Harassment named on screen, where the user reads it.
Visual description
A four-segment semicircular diagram labelled “4 tensions” connects by lines to the four statements below. Shades of violet and navy distinguish the segments. Segment sizes are illustrative, not numerical measurements.
Slide 22 of 29: Accessibility built into the interface
Accessibility.
Interface provisions stated in the presentation
- Roles and labels: Every control is labelled for VoiceOver.
- Dynamic announcements: New detections are announced out loud.
- Reduce motion: The pulse stops and the radar simplifies, following the iOS system setting.
- Haptic feedback: Vibration as a second channel when the screen is out of view.
- Contrast 4.5:1+: Checked on the core screens: idle, scanning, detected, about.
- Legible type: Urbanist in four weights, applied across every component.
Visual description
Six white cards are arranged in two rows of three on a pale background. Each contains a violet icon, a feature heading and explanatory text. The slide describes provisions and contrast checks, not a formal WCAG conformance certification.
Slide 23 of 29: AI tools, by phase
Attribution. Column headings: Phase, AI tools used, We decided.
Table rows
- Empathy. AI tools used: Claude, Perplexity, NotebookLM. We decided: Problem statement and ethical framing.
- Ideation. AI tools used: Claude, ChatGPT. We decided: NearLens, over every pun we generated.
- Prototyping. AI tools used: Replit Agent, Google Stitch. We decided: Architecture, palette, privacy defaults.
- Evaluation. AI tools used: Attention Insight. We decided: What the heatmaps actually meant.
Visual description
The current replacement slide uses a violet background. Four horizontal rows align phase icons and names on the left, tool-name pills in the middle and a darker right-hand column for human decisions. There is no fifth row or Userbrain entry in this version.
Slide 24 of 29: Built in a month, in 410 commits
Build. The headline total is 410 commits; the chart specifically breaks down the 328 commits of Phase 1.
Where the 328 commits of Phase 1 went
- Visual refinement: 38%.
- Features: 18%.
- Identity and branding: 14%.
- Layout and composition: 10%.
- Accessibility: 5%.
- Config, docs and fixes: 15%.
What was built
- BLE detection: Passive Bluetooth reading, manufacturer by ID.
- Animation: 60 fps radar, idle pulse.
- Accessibility: 4.5:1 contrast, screen reader support, no forced animation.
- Data and privacy: On-device only. No server, no tracking.
- Publication: App Store, nearlens.app, Web build (Expo).
Visual description
Six violet horizontal bars on the left encode the printed percentages, with visual refinement longest and accessibility shortest. Five icon-led capability descriptions appear on the right.
Slide 25 of 29: Field observations in public space
Public testing. Six observations, four cities. 1,564 devices counted where totals were logged.
Locations and reported results
- TD Place Arena, Ottawa, Canada: 15 detections. Crowd approximately 7,000 people. Window under 3 hours. Company ID versus name-pattern matches: 11 versus 4.
- Meridian Theatres, Ottawa, Canada: 0 detections. 823 BLE devices seen. 0 smart glasses. Likely cause stated on the slide: norms and demographics.
- International airports: YOW, Ottawa, 1 detection; YYZ, Toronto, 0; SFO, San Francisco, 0. Airport group total: 1 detection.
- Washington DC, USA: 1 detection. 741 devices, 38 scans, −87 dBm, over 15 metres. Identified in the slide as a VR headset, Quest 3.
Visual description
Four location cards combine photographs with results: the arena and theatre across the top, airports and Washington DC below. The printed counts are field observations, not a controlled accuracy benchmark; a VR headset is included among the reported matches.
Slide 26 of 29: Nowhere is this a local problem
Early impact. Where it landed.
Printed reach figures
- 52 countries. Top: US, Canada, Germany, UK.
- 6 continents. North America to Oceania.
- $0 spent to reach them. 100% organic.
Press and recognition
- Press in seven countries picked the app up on their own: Canada, the USA, France, Germany, Italy, China and Denmark.
- Visible publication marks: CBC, betaKit, AppDeveloper Magazine, Tencent, Innovations.fr, TechNX, Bay Street Wire, JdT, Mobilelabs and INOVATIA TECH.
- FPD, Future Product Days. Honorable Mention: Pushing the AI Frontier.
- Source printed on the slide: App Store Connect, March to September 2026.
Visual description
Three large reach figures occupy the left and centre on a dark background. Press logos run below. A white three-dimensional globe surrounded by violet concentric rings appears on the right. The globe is decorative, not a country-by-country map.
Slide 27 of 29: Next steps
Roadmap. From the first build to version 2.4.2, and what still needs testing.
Roadmap stages
- 1. NearLens v1: Bluetooth device awareness and local history. Six field observations. Shipped, launch. iOS app, 30 days.
- 2. NearLens 2: Privacy Intelligence, Live Activity, and the interface rebuilt on Signals. Shipped, version 2. Releases 2.0 to 2.4.1.
- 3. Privacy signals: PMSG recording signals, demo mode, and new scan-interval controls. Released, October 2026. Version 2.4.2.
- 4. Test with real hardware: Accuracy, latency and battery, in three countries. Next: field validation.
Visual description
A horizontal line links four numbered cards. The first three have solid violet nodes to indicate shipped work; the fourth has a subdued outlined node to indicate planned validation.
Slide 28 of 29: The problem is still open.
Much more to do.
NearLens treats a symptom. The structural problem is the consent asymmetry in platform design.
What it demonstrates
- Passive detection is technically feasible, local-only, and free.
Where UX stops short
- Consent flows reach only the wearer. The people in frame were never part of the contract.
What it demands
- A response at the platform level.
Visual description
The conclusion is set in white on a violet background. On the right, a man wearing camera glasses raises a hand to the frame; concentric arcs behind his head suggest a transmitted signal.
Slide 29 of 29: Thank you
Thank you.
nearlens.app
Marcos Rezende.
Link
- Marcos Rezende's name links to marcosrezende.com.
Visual description
A small NearLens icon is centred above the closing message, website and underlined author name on a dark navy background.