Samuel J Lee

Ludus

|www.glaze.app/app/ludus-4RDpwn|

GlazeAnthropicOpenAIGemini

2026

Multi-provider AI tutor that turns any topic interest or content upload into structured, mastery-based courses.

Problem

Most AI learning tools dump chat on the learner. There is no curriculum, no evidence of mastery, and no way to choose which model actually teaches you well.

Solution

Ludus generates organised courses from a single interest or uploaded materials, unlocks lessons as you progress, and pairs each lesson with a context-aware tutor. Projects and exercises create proof of understanding; collector coins track momentum. Tutors and course generation run across Glaze, OpenAI, Anthropic, Google, and Ollama.

Thinking

The product bet is structure over chat: AI should author a learning path, not just answer questions. Multi-provider support treats models as interchangeable engines so the pedagogy stays constant while the underlying LLM can change.

Ephello

|

Next.jsVercelCursorClaude

2026

Community-health platform for councils

Problem

Councils relate to residents through newsletters, siloed CRMs or at worst, not at all. Signals of neighbourhood strain arrive late, residents repeat their story in multiple places, a true measure of community cohesion and health is difficult to understand.

Solution

Ephello turns community health into an operational system: listen for neighbourhood signals, engage with timely outreach, support shared case trails, and measure ward-level health. AI assists drafting and coordination so staff move faster without losing continuity.

Thinking

Treat communities as living systems with a heartbeat you can monitor.

Berry

|

Next.jsCursorVercelCloudflareOpenAIGoogle AI

2026

Shared university notes with an AI companion that summarises, quizzes, and explains.

Problem

Lecture notes scatter across drives and group chats. Studying alone means re-reading without feedback, and good notes rarely compound across a cohort.

Solution

Berry lets students join courses, share and discover notes, and study with an AI companion that summarises, quizzes, and explains concepts grounded in those notes — built on Next.js with OpenAI and Google AI.

Thinking

The moat is the shared corpus, not the chat widget. Ground the model in classmates' notes so explanations stay course-specific and the network effect improves study quality.

BOS

|

TauriReactViteTailwindProseMirrorSQLiteComposioAnthropicOpenAIGoogle AI

2026

Local-first daily OS that aggregates tools and LLMs on-device with strict data sovereignty.

Problem

Knowledge work is fragmented across SaaS apps, and AI features usually mean shipping private context to someone else's cloud.

Solution

BOS is a Tauri desktop shell with a ProseMirror workspace, SQLite storage, and Composio integrations. Multiple LLM providers (Anthropic, OpenAI, Google, Ollama) sit behind one local interface so daily ops stay private by default.

Thinking

Build the aggregator where the data already lives — on the machine. Treat cloud LLMs as optional compute, not the system of record, so privacy and continuity survive vendor churn.

Brix

|

Next.jsOpenAIiMessageNeon

2026

AI-managed repair syndication that gives small landlords institutional buying power via iMessage.

Problem

Small landlords pay retail rates for repairs because demand is fragmented and coordinating trades is high-friction admin.

Solution

Landlords text maintenance requests over iMessage. Brix aggregates localised demand, uses OpenAI to structure and match jobs, and unlocks group discounts from tradespeople — collective buying without becoming a property manager.

Thinking

Syndication only works if intake is almost zero effort. iMessage is the capture surface; AI is the ops layer that turns messy texts into matchable, aggregatable demand.

Echo

|

iMessageSendblueClaude API

2026

Zero-learning-curve maintenance logging for manufacturing teams via iMessage + Claude.

Problem

Frontline manufacturing teams will not adopt heavy enterprise CMMS tools. Issues get lost in verbal handoffs, WhatsApp chaos, or never logged at all.

Solution

Workers report equipment issues by text or voice note in iMessage. Claude structures those messages into centralised operational logs — no new app to learn, no training programme.

Thinking

Meet operators in the channel they already use. Put AI on the structuring layer (taxonomy, severity, searchable logs) rather than forcing a new UI onto the shop floor.

Voiceprint

|

ClaudeNotion

2026

Autonomous voice-profile system that learns a writer's style and keeps generating on-brand copy.

Problem

AI writing defaults to generic tone. Creators waste time restating voice rules, and outputs drift away from how they actually sound.

Solution

Voiceprint analyses existing writing to build platform-specific voice profiles — tone, style, signature phrases — then runs a continuous learning loop so new drafts stay authentic to the creator.

Thinking

Voice is a living system, not a one-shot prompt. Encode it as structured profiles and keep feeding real outputs back so the model converges on the person, not a stereotype.

Rememberall

|

CursorSwift

2026

Mac clipboard and screenshot history, visually grouped so you can find and paste anything fast.

Problem

Clipboard history is either invisible or trapped in clunky utilities. Screenshots and copied text pile up with no useful organisation.

Solution

Rememberall saves clipboard items and screenshots into a visual history, grouped by type, app, and custom categories — search, find, and paste in seconds.

Thinking

Memory tools fail when recall is slower than redoing the work. Optimise for visual scan and paste speed, not archival completeness.

Proppy

|

Next.jsVercelSupabase

2026

Side-by-side property comparison with structured, AI-driven viewing checklists for homebuyers.

Problem

Comparing homes is cognitively heavy — notes scatter, gut feel dominates, and viewing assessments are inconsistent across properties.

Solution

Proppy centralises property data into a clean comparison dashboard with structured viewing checklists, reducing ambiguity when evaluating multiple homes.

Thinking

Homebuying decisions fail from unstructured evidence. Force a shared checklist schema so comparison is apples-to-apples, not nostalgia vs floorplan.

Eros

|

CursorFirecrawlTauri

2026

Local-first research CRM — one-click Chrome capture, Firecrawl enrichment, Tauri desktop browse.

Problem

Company research lives in tabs, Notion dumps, and half-finished spreadsheets. Enrichment tools are cloud-heavy and context gets lost between sessions.

Solution

Eros saves companies from Chrome with one click, enriches them with Firecrawl, and keeps everything browseable in a local Tauri app — research that stays on your machine.

Thinking

Agentic scraping is most useful when the CRM is local-first. Capture frictionlessly in the browser; enrich asynchronously; never make the research graph hostage to a SaaS login.

Fairwealth

|

ExpoClaudeNeon

2026

AI-powered wealth management for household and individual finances.

Problem

Personal finance tools show charts but rarely help people decide what to do next in plain language across accounts, goals, and messy real life.

Solution

Fairwealth is an Expo/React Native app backed by Neon that uses AI to interpret household finances and surface actionable guidance — not just dashboards.

Thinking

Wealth UX fails when it assumes spreadsheet literacy. Put a conversational AI layer on structured finance data so recommendations feel personal and executable.

Fika

|

SwiftUIMapKitCursorClaude Code

2026

A quiet map of time shared with people you care about — coffee, walks, dinners that would otherwise fade.

Problem

Time with people we care about dissolves into camera rolls and vague memory. Months later you cannot say when you last saw someone properly.

Solution

Fika maps the moments you actually shared, becoming a curated personal story of real life lived together — built with SwiftUI, MapKit, and Claude Code.

Thinking

Relationship memory is spatial and temporal, not a feed. Make presence legible on a map so the product rewards noticing people, not posting about them.

Glayze

|

Claude CodeSupabaseSwift

2026

Native iOS dessert-map app with on-the-spot QR loyalty for independent shops.

Problem

Dessert discovery is fragmented, and independent shops lack simple loyalty that works at the till without another tablet workflow.

Solution

Glayze maps donuts, gelato, bakeries and more; after purchase, customers show a QR code and the merchant approves points on the spot — redeem perks as you explore.

Thinking

Loyalty only sticks if capture is instantaneous for both sides. Map discovery drives footfall; QR approval keeps merchants in control without new hardware.

Kwesty

|

SwiftSupabaseReve

2026

Kids' routine timer that turns chores into time-trial challenges with badges.

Problem

Getting children through daily routines is friction for parents and boring for kids — timers feel like nagging, not play.

Solution

Kwesty gamifies routines: beat your previous time on an activity and earn badges, turning mundane tasks into challenges backed by Swift and Supabase.

Thinking

Motivation for kids is personal bests, not parental dashboards. Design for the child's loop first; parent oversight is secondary infrastructure.