{"id":51475375,"url":"https://github.com/asmuelle/kinsight","last_synced_at":"2026-07-06T20:30:34.268Z","repository":{"id":366029001,"uuid":"1266479866","full_name":"asmuelle/kinsight","owner":"asmuelle","description":"Turn any spare phone into an all-local fall \u0026 anomaly monitor for aging parents — only text alerts ever leave the device; raw video never exists off it.","archived":false,"fork":false,"pushed_at":"2026-06-19T23:13:38.000Z","size":253,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-20T01:09:08.541Z","etag":null,"topics":["android","computer-vision","edge-ai","elder-care","ios","on-device-inference","privacy"],"latest_commit_sha":null,"homepage":"https://asmuelle.github.io/kinsight/","language":"Kotlin","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/asmuelle.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":"AGENTS.md","dco":null,"cla":null}},"created_at":"2026-06-11T16:51:05.000Z","updated_at":"2026-06-19T23:36:41.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/asmuelle/kinsight","commit_stats":null,"previous_names":["asmuelle/kinsight"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/asmuelle/kinsight","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/asmuelle%2Fkinsight","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/asmuelle%2Fkinsight/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/asmuelle%2Fkinsight/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/asmuelle%2Fkinsight/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/asmuelle","download_url":"https://codeload.github.com/asmuelle/kinsight/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/asmuelle%2Fkinsight/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":35205739,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-07-06T02:00:07.184Z","response_time":106,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["android","computer-vision","edge-ai","elder-care","ios","on-device-inference","privacy"],"created_at":"2026-07-06T20:30:33.472Z","updated_at":"2026-07-06T20:30:34.234Z","avatar_url":"https://github.com/asmuelle.png","language":"Kotlin","funding_links":[],"categories":[],"sub_categories":[],"readme":"# KinSight\n\n[![CI](https://github.com/asmuelle/kinsight/actions/workflows/ci.yml/badge.svg)](https://github.com/asmuelle/kinsight/actions/workflows/ci.yml)\n\n\u003e Turn any spare phone into an all-local fall and anomaly monitor for aging parents — only text alerts ever leave the device; raw video never exists off it.\n\n**Category:** Edge AI / on-device inference (iOS + Android) · **Status:** 🟡 Runner-up — viable with caveats\n\n## Scorecard\n\n| Metric | Score |\n|---|---|\n| Rank (of 9 finalists) | #6 |\n| Combined score | 5.2 |\n| Monetization potential (1-10) | 7 |\n| Feasibility (1-10) | 5 |\n| Edge AI structurally essential | Yes |\n| Skeptic verdict | weakened |\n\n## Concept\n\nTurn any spare phone into an all-local fall and anomaly monitor for aging parents — only text alerts ever leave the device; raw video never exists off it.\n\n## Target User \u0026 Payer\n\nAdult children (45-65) of independently-living elders — already paying $30-60/mo for medical-alert pendants their parents refuse to wear, and rejecting in-home cloud cameras over privacy. Classic buyer-is-not-the-user care economics.\n\n## Why Edge AI Is Structural (not decoration)\n\nContinuous Vision/MediaPipe pose and motion analysis on a repurposed wall-mounted phone detects falls, prolonged immobility, and routine deviations (missed kitchen visits = missed meals); AFM 3 image input or Gemma 3n vision locally verifies candidate anomaly frames to crush false positives before alerting; only a text alert and activity summary transmit, verifiable by network audit. Essential three ways: 24/7 cloud video streaming is the #1 documented adoption objection, is bandwidth/cost-prohibitive at always-on scale, and fall alerts need second-level latency independent of uplink.\n\n## Why Now (2026 timing)\n\nAI elder care projected at $208B by 2032 (25% CAGR) with camera privacy as the named top blocker; the EU AI Act (Aug 2026) makes cloud-video elder monitoring legally radioactive; AFM 3 on-device vision and Gemma 3n vision just made local anomaly verification feasible on 2-3-year-old donor phones.\n\n## Proposed Monetization\n\n$19.99/mo or $149/yr per household billed to the adult child, priced below the medical-alert services it replaces; multi-room/multi-parent family plan $249/yr. Hardware cost zero (donor phone); ~100% gross margin since no video infrastructure exists at all.\n\n## Competition \u0026 Gap\n\nNest/Ring-style cameras and elder platforms stream video to servers — the exact objection blocking adoption; wearable pendants (Life Alert) fail because elders don't wear them. No incumbent can match 'video never leaves the home' without an architectural rebuild.\n\n---\n\n# Evaluation (multi-agent adversarial review)\n\n## Monetization Analysis — score 7/10\n\nThe payer, price point, and pain are all verified: adult children already pay $33-46/mo for medical-alert services (Medical Guardian $32.95-37.95/mo, Kami Fall Detect $45/mo plus $99 hardware), the elderly-monitoring market is real ($2.5-6.4B for the device/monitoring slice, 11-12% CAGR — the pitch's $208B figure is the inflated all-of-AI-elder-care number, but the addressable slice is still large), and VCs have validated fall detection specifically with $123.6M into SafelyYou and $296M into Vayyar at a $1B valuation. The buyer-is-not-user dynamic supports premium recurring pricing and low price sensitivity. The privacy wedge is documented — AltumView, Butlr, and Vayyar all market 'no video leaves the home' positioning, confirming the objection is real. However, three things cap the score below 8: (1) price anchoring risk — AlfredCamera proves the exact donor-phone mechanic at 90M users but monetizes at $2.49-5.99/mo and only ~$200K/mo estimated US Google Play revenue, so 'an app on an old phone' is anchored at 1/4 the proposed price and KinSight must win the comparison against medical-alert services, not camera apps; (2) the privacy objection is being attacked by radar/thermal competitors whose 'no camera at all' arguably beats 'camera that promises not to upload' in elder perception, and AltumView already ships privacy-preserving fall detection with free basic features; (3) safety-critical liability and reliability on aging donor phones (battery swelling, OS killing the app, no 24/7 monitored response center at 3am) creates support cost and brand-catastrophe risk that pure-software margins understate, plus churn is structurally event-driven (parent moves to care or dies), bounding LTV at roughly 2-3 years like the medical-alert industry. Net: solid, evidence-backed niche with a plausible wedge into a proven-willing-to-pay market, but unproven at the proposed price for the proposed form factor — a 7.\n\n## Recommended Revenue Model\n\nKeep the proposed $19.99/mo, but anchor and message exclusively against medical-alert services ($32.95-45/mo verified), never against camera apps. Recommended structure: 30-day free trial (trust is the conversion barrier in this category), then $19.99/mo or $149/yr single household; $249/yr family plan (multi-room/multi-parent, 2-4 donor phones) as the primary push since annual prepay matches the adult-child gifting moment (holidays, post-hospitalization). Add a +$9.99/mo optional monitored-response tier via a RapidSOS-style partnership (NOMO already does this at $19.99/mo total) to close the '3am and the adult child is asleep' gap — without it KinSight is alerting-only and loses head-to-head against Life Alert-class services on the scariest scenario. Unit economics: ~100% gross margin is roughly right (no video infrastructure; on-device inference), so at $149/yr effective ARPU and 2.5-year event-driven lifetime, LTV ~$370/household; CAC via caregiver-content SEO and Facebook 45-65 targeting plausibly $60-120, giving healthy 3-6x LTV/CAC. Realistic trajectory: 5,000 paying households = ~$750K ARR (strong indie outcome); 25,000 = ~$3.7M ARR (venture-viable, comparable to where SafelyYou was at Series A). Avoid lifetime pricing; the event-driven churn already caps LTV.\n\n## Market Evidence (live web research, June 2026)\n\nVerified via web search (June 2026): Medical-alert incumbents charge $32.95-37.95/mo (Medical Guardian) up to $125/mo (Rest Assured, plus $350 install); Kami Fall Detect camera is $99 hardware + $45/mo; NOMO Smart Care is $19.99/mo with RapidSOS emergency response — confirming the proposed $19.99/mo sits credibly below incumbents. Market size: elderly monitors market projected at $6.44B by 2029 (11.4% CAGR, Business Research Company); smart elderly monitoring systems $2.5B in 2025 growing to $6.8B by 2034 (Verified Market Reports); the pitch's $208B figure corresponds to the much broader AI-in-elder-care aggregate (InsightAce: $56.78B in 2025 to $387.5B by 2035) — directionally large but the monitoring slice is single-digit billions. Investor validation: SafelyYou raised $123.6M total for AI camera fall detection (B2B senior-living focus, 450+ communities); Vayyar raised $296M at a $1B valuation for radar fall detection — but Vayyar's reported revenue was negligible (~£13.2K as of FY2023 per Tracxn), a caution that privacy-first fall-detection hardware has badly lagged its funding. Closest mechanical comp AlfredCamera: 90M+ users repurposing old phones as cameras, premium at $2.49-5.99/mo, ~$200K/mo estimated US Google Play revenue (Sensor Tower) — proves the donor-phone mechanic at scale but shows weak per-user monetization at commodity-camera positioning. Privacy objection confirmed real: AltumView Sentinare and Butlr both market privacy-preserving (videoless/no-camera) elder fall detection as their core pitch.\n\n## Comparables\n\n- AlfredCamera (donor-phone camera app): 90M+ users; premium $2.49/mo annual or $5.99/mo; ~$200K/mo estimated US Google Play revenue per Sensor Tower — proves the mechanic, anchors price low\n- Medical Guardian (medical alert): $32.95-37.95/mo, $149.95 equipment on home cellular — the incumbent KinSight prices under\n- Kami Fall Detect (AI fall camera): $99.99 hardware + $45/mo subscription with emergency response\n- NOMO Smart Care (no-camera home medical alert): $19.99/mo with RapidSOS 24/7 response — exact price point KinSight proposes, with a response center included\n- CarePredict @Home (D2C wearable elder monitoring): $449.99 kit + $69.99/mo — shows premium D2C tolerance; largely pivoted to B2B senior living\n- AltumView Sentinare (privacy-preserving stick-figure fall sensor): ~$200 hardware, basic features free, optional subscription — closest privacy-positioned competitor\n- SafelyYou (AI fall detection cameras, B2B): $123.6M raised, deployed in 450+ senior-living communities\n- Vayyar (radar fall detection): $296M raised, $1B valuation 2022, but ~£13.2K reported revenue FY2023 — cautionary comp on monetizing privacy-first fall tech\n- Rest Assured (monitored elder care): $350 install + $125/mo — top of the price umbrella\n\n## Adversarial Review — strongest case AGAINST (verdict: weakened)\n\nThe headline promise — reliable fall detection — fails exactly when and where elders actually fall. Falls cluster at night (bed-to-bathroom transits) and in bathrooms; a donor phone has no IR illuminator so its RGB camera is blind in the dark, and cameras are socially unacceptable in bathrooms, so the highest-risk windows are structurally uncovered. Meanwhile the privacy-first fall market is already owned by mmWave radar (Vayyar Care, SafelyYou-class edge systems): no camera at all, works in total darkness, deployable in bathrooms — 'private camera' loses to 'no camera' on KinSight's own founding objection, falsifying the 'no incumbent can match us' claim. The donor-phone economics contradict the model requirements: AFM image input (new at WWDC26) needs Apple Intelligence-class hardware (iPhone 15 Pro+/A17 Pro), and Gemma 3n vision needs ~3GB+ RAM on a 2022+ SoC, while real drawer phones are iPhone 11-13s and low-RAM Androids. A ~3B-class VLM verifier cannot reliably distinguish 'fallen on floor' from 'lying on couch' or 'kneeling'; tuning it to crush false positives suppresses true positives, and a missed fall in a product priced as a Life Alert replacement is catastrophic legal exposure (with EU AI Act high-risk classification plausibly applying to elder-safety AI regardless of on-device inference — that claimed moat cuts both ways). Finally, 24/7 camera+NPU on a wall-mounted consumer phone is an uptime nightmare: iOS forbids true background camera capture (kiosk/Guided Access hacks only), thermal throttling, swelling aged batteries at 100% charge duty cycle, OS auto-updates and post-outage reboots that only the unsupervised elder can fix. The alert path still requires uplink, so 'latency independent of uplink' only holds for a local siren.\n\n## Recommended Tech Stack\n\nAndroid monitor device (primary — only platform where 24/7 camera is tractable): CameraX + foreground service with FOREGROUND_SERVICE_TYPE_CAMERA; cheap luma frame-diff motion gate; MediaPipe Pose Landmarker (LiteRT, GPU/NNAPI delegate) at 5-10fps when motion present; custom temporal fall/immobility classifier (1D-CNN/GRU over keypoint sequences) exported to LiteRT; Gemma 3n E2B (.litertlm) via MediaPipe LLM Inference API for candidate-frame verification on \u003e=4GB-RAM devices, deterministic heuristics below that; local siren + FCM for alerts; watchdog with auto-relaunch, OEM charge limiting, thermal governor dropping to ~1fps idle. iOS monitor device (secondary, newer hardware only): kiosk-style always-foreground app under Guided Access; AVFoundation capture + Vision VNDetectHumanBodyPoseRequest/3D; Core ML temporal classifier (coremltools); FoundationModels image input (iOS 26.x+, A17 Pro+) or quantized FastVLM via Core ML/MLX as verifier; APNs alerts. Caregiver app (iOS/Android): push alerts plus a daily routine digest generated on the caregiver's own phone from event metadata only — Apple Foundation Models text generation on iOS, ML Kit GenAI summarization (Gemini Nano/AICore) on Android. Infra: thin push-relay only (no media path ever), E2E-encrypted event metadata. Product repositioning baked into stack: ship daytime activity/routine-deviation monitoring as the primary claim; fall detection as best-effort wellness feature, optionally paired with a $15 plug-in IR illuminator for night coverage.\n\n---\n\n*Generated 2026-06-10 from a multi-agent research pipeline: 5 live-web research agents (Apple/Android platform state, market data, consumer trends, competitive landscape), 3-lens ideation, ruthless shortlist, then per-candidate monetization analyst + adversarial skeptic. Market figures are agent-researched estimates — verify before committing capital.*\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fasmuelle%2Fkinsight","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fasmuelle%2Fkinsight","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fasmuelle%2Fkinsight/lists"}