Music Memory AI

The description states that Music Memory AI allows users to search for songs using incomplete or vague memory cues such as:

• Approximate or inaccurate lyrics
• Movie, series, or advertisement context
• Mood, tempo, genre, or instruments
• Singer or voice characteristics
• Scene details

It returns ranked results with explanations, evidence categories, and YouTube links. Users can refine searches, confirm or reject candidates, and save discoveries.

The system uses three retrieval strategies:

1. Direct — for distinctive phrases or title fragments.
2. AI-assisted — using GPT-5.6 to extract clues and propose diversified queries.
3. Smart — chooses between direct and AI-assisted based on input.

GPT-5.6 is used only for structured clue extraction and query generation, not for ranking or inserting unsupported metadata.

YouTube provides candidates, and TypeScript code normalizes and ranks them deterministically, selecting the best representative recording after aggregation of evidence across versions.

Frontend: React, TypeScript, Vite on Cloudflare Pages

Backend: Hono API on Cloudflare Workers

Authentication & saved songs: Supabase

Tools used: Codex, OpenAI, YouTube, MediaRecorder, Zod

Inference The product is a search engine for music that attempts to bridge memory gaps using AI and external data sources.

The author states the core promise:

“Find the song you almost remember from incomplete lyrics, descriptions, context, era, mood, or voice clues.”

This positioning targets people who struggle to recall exact song titles or lyrics but have a sense of what they're looking for — a common problem in music discovery.

There is no evidence of prior versions or iterative positioning. The project is described as a single submission to a hackathon.

Inference The product positions itself as a niche tool for memory-based music search, not a general-purpose music platform or streaming service.

The description does not name specific customer segments or personas. It implies the user base includes people who:

• Have partial memories of songs
• Are trying to identify music from context (e.g., movie scenes)
• Want to explore music based on mood, genre, or voice

It is unclear if the target audience is casual users, music enthusiasts, or professionals like DJs or content creators.

Inference The ICP likely includes individuals with vague musical memory and a need for discovery tools — but no explicit segmentation or targeting data is provided.

There is no evidence of pricing, monetization, or business model in the description. The project is described as a hackathon submission without any indication of commercial intent or revenue streams.

Inference No business model or pricing structure is evidenced; this remains unknown.

The system uses:

• GPT-5.6 for clue extraction and query generation (not ranking)
• Deterministic code to resolve identity, aggregate evidence, and rank results
• External grounding via YouTube and other providers
• Cloudflare Workers + Pages for deployment
• Supabase for authentication and saved songs
• React + TypeScript frontend

Codex was used throughout development as an engineering collaborator.

The system passes:

• Strict TypeScript checking
• ESLint
• 240 automated tests
• Production builds for Worker and frontend
• All 24 benchmark case/pipeline runs with perfect Recall@1, Recall@3, Recall@5, and MRR

Inference The technical architecture is well-defined and tested in development. However, no live performance or scalability data is provided.

The project is described as a hackathon submission (OpenAI 2026). No evidence of:

• Users
• Revenue
• Customers
• Product usage metrics
• Market traction

It has not been demonstrated beyond the author’s own account and recorded test cases.

Inference There is no evidence of traction or maturity beyond a prototype built in a short time.

The description does not mention competitors. It implies the product addresses a gap in traditional music search, where users must know precise terms.

It is unclear whether similar tools exist (e.g., Shazam, SoundHound, AI-based music identification services), but no such references are made.

Inference No competitive landscape is described or evidenced.

• Unverified claims: The product is presented as a working solution, but there is no independent validation.
• No user feedback or adoption: No evidence of real-world usage or customer data.
• Limited scope: The system only works with YouTube and external providers; it does not appear to include full music libraries or APIs.
• AI dependency: Reliance on GPT-5.6 for clue extraction, which may be inconsistent or unreliable in live use.
• Hackathon origin: The project is a one-off submission, not a developed product.

Inference Risks include lack of real-world testing, unproven scalability, and no commercial viability.

1. What are the actual limitations of the system in live use?
2. How does it handle edge cases or ambiguous inputs?
3. Has anyone used this beyond the development phase?
4. Are there plans to integrate with music libraries or APIs beyond YouTube?
5. How is the ranking algorithm validated in practice?
6. What are the long-term plans for monetization or product evolution?