AI Enterprise SolutionsGPT & Claude

Obsidian

Target users for FamCanvas AI span three primary groups: long-distance family members like students studying abroad, busy working professionals, and multi-generational family heads. Students living away from home need an effortless, low-friction way to share spontaneous micro-moments—such as a photo of lunch or a quick audio note—without the pressure of scheduling long phone calls. Exhausted working parents returning home from demanding jobs need a comforting, centralized space to catch up on what their family shared during the day and leave quick reactions. Meanwhile, older family members require a clean, highly visual digital space where shared memories are displayed clearly without navigating complex social media feeds. The underlying problem is that modern communication tools keep families connected transactionally, but fail to maintain emotional intimacy or preserve collective memories. Today, families rely heavily on instant messaging apps where meaningful photos, personal thoughts, and voice notes quickly get buried under everyday administrative noise and logistical clutter. Alternatively, cloud storage platforms automatically back up thousands of photos into passive digital archives that families rarely revisit or reflect on together. This results in an emotional disconnect where long-distance relatives default to brief, superficial check-ins, causing everyday micro-moments to fade away and leaving years of shared family history fragmented across scattered devices. FamCanvas AI bridges this gap by replacing passive storage and noisy group chats with an interactive digital living space. By combining a real-time shared whiteboard with multimodal generative AI, it gives families a casual environment to post daily thoughts and photos, while automatically turning those everyday interactions into a rich, narrated family memory heritage.

GitHub

Preliminary score

54.5/ 100Earlyaverage of 2 judges

Value for the user14.0 / 25
A specific problem and a clear outcome from solving it.
Prototype and use of AI14.5 / 30
A working core scenario, and what AI actually contributes to the result.
Quality testing11.5 / 20
Tests, examples of failures, and comparison with the current approach.
Feasibility9.5 / 15
Data requirements, running costs and a clear next step.
Originality5.0 / 10
A meaningful difference from common solutions.

What the judges wrote

Claude Opus 5.5Preliminary50/100
Strengths
Your repository shows serious engineering for a one-day build: Supabase row-level security isolating families, atomic monthly quotas with refunds for failed generations, signed Stripe webhooks, and a timezone-aware reminder cron with idempotency. Your README documents a database test suite on PGlite covering family isolation, roles, quotas and invitations, plus Playwright browser tests. Your write-up names specific failures you hit, such as 413 payload errors from uncompressed canvas images and flaky async audio-upload tests, and explains how you fixed each one. The offline demo mode with a reset button is a thoughtful way to make the presentation reliable.
To improve
Your AI core, GPT-4o vision storybooks plus TTS narration, is not shown working: the demo uses a deterministic sample story, and the README says live services were never tested. That also conflicts with your claim of observing over-dramatic narratives, so include real generated stories, including bad ones, as evidence. Your materials disagree with each other on the model (GPT-6 Astra vs GPT-4o), the Next.js version (14 vs 16) and pricing ($3.99/$7.99 vs $6.99/$59), and these details need to match. You entered the Enterprise track, but the 'Business Corporates' audience has no workflow or adoption metric; either define one or pick a single specific user and a measurable outcome, such as weekly revisits or memories shared per family. Add a rough cost per story, for example vision tokens plus TTS minutes, and set it against your subscription price.
GPT-6 AstraPreliminary59/100
Strengths
Your focus on distant relatives, busy parents, and older family members identifies concrete frustrations with noisy chats and forgotten photo archives. Your shared canvas, narrated storybooks, and reminders form a coherent family-memory experience. Your README clearly distinguishes deterministic demo stories from live AI, documents substantial permission and quota safeguards, and acknowledges testing limitations alongside specific engineering failures.
To improve
Demonstrate an authenticated two-user workflow through real story generation and narration, and reconcile the conflicting model, framework, and pricing descriptions. Extend your documented tests with recorded results, narrative-fidelity failure cases, and comparison against WhatsApp or shared photo albums. Quantify vision, narration, storage, and reminder costs per active family to validate your subscription pricing. Your enterprise-track fit remains unproven: identify a corporate workflow and measurable business outcome, or clarify the consumer positioning.

The team

Alp Eren ShahbazliLead
product
Ismayil Ismayilzada
business
Aliakbar Rzazada
developer

Disclosure

Required by the rules. Judges verify it against the repository.

Models: GPT-6 Astra
Data: User-uploaded photos and notes, synthetic family activity logs, Unsplash public sample images
Components: Next.js 14, React, TypeScript, Tailwind CSS, Supabase (Auth, Realtime DB, Storage), Fabric.js, Stripe API, Vercel
Built during the event: Confirmed

Quality testing

Building FamCanvas AI revealed several real-world engineering and adoption challenges that we had to confront head-on. From a product perspective, relying heavily on GPT-4o Vision for visual analysis and OpenAI TTS for speech narration introduces noticeable operational costs that scale quickly as families post high-resolution photos and voice notes. Furthermore, overcoming user inertia remains a subtle challenge, as convincing busy parents or older relatives to adopt a new digital space over established routines like WhatsApp requires intuitive design and gentle, non-intrusive notification loops. We also observed edge cases where thin image contexts occasionally led the AI storyteller to generate slightly over-dramatic narratives, alongside the infrastructure demand of syncing rich real-time canvas states across devices. Under the pressure of the build, we made a few genuine technical missteps that tested our resilience. We initially attempted to send raw, uncompressed canvas images directly to the vision endpoint, which triggered 413 payload overflow errors and halted our pipeline. We also lost valuable time troubleshooting Next.js 16 build failures caused by explicit runtime flags conflicting with Cache Components, while our automated Playwright end-to-end browser tests repeatedly timed out while waiting for asynchronous audio file uploads to settle on the interactive whiteboard. Far from undermining the project, exposing these flaws ultimately makes FamCanvas AI significantly stronger and more viable. Hitting the image payload limit forced us to implement strict 32KB payload capping and low-detail image parameters, while the Playwright timeouts compelled us to replace fragile sleep delays with robust, reactive state assertions. Confronting these exact failure modes during testing gives us an honest, mature defense for judge Q&A, proving that we understand our architecture's real-world constraints and have already laid out a clear engineering roadmap for cost optimization, batch processing, and scalable family memory management.

Worth 20 of the 100 points.

Project ownership

The team keeps the rights to this project. Judges were given access for verification only.