Training Data for
AI Video Edit Model
We're building an AI-powered video editing assistant inside our robot companion platform. We need a music curator to build a structured, tagged library of tracks that will train our model to understand music-to-mood-to-edit relationships.
01Background
MDR (Mondo Robotics) is a Shenzhen-based robotics company building LZ, a consumer companion robot with a high-quality onboard camera. The robot's software platform handles everything from camera control to AI-assisted content creation.
A core feature of the platform is its AI video editing engine — an automated system that takes raw footage captured by LZ and produces polished, emotionally resonant edits. Music is the backbone of this editing intelligence. The model needs to learn how different musical qualities (tempo, energy, mood, instrumentation) map to editing decisions (cut pace, transition style, color grading mood, motion dynamics).
Your role: Curate, source, and tag a diverse music library that becomes the training foundation for this AI. You're not picking music for commercials or ads — you're picking music that real people would use on their own social media. The kind of tracks that feel organic, natural, and comfortable. The songs that go viral because they just fit — not because someone paid for placement.
02Brand Context
Understanding MDR's brand personality is essential because the music you select will shape the emotional palette of every video LZ produces. The brand sits at the intersection of:
- Curiosity — exploration, wonder, discovery. The primary brand value.
- Vitality — energy, movement, aliveness. The secondary driver.
- Human Connection — warmth, intimacy, belonging. The emotional anchor.
Think of it as: SpaceX's ambition meets Apple's craft meets the soul of a Saturday morning adventure with someone you love.
Tone Keywords
What We're NOT
03Scope of Work
Track Curation (Core Deliverable)
Source and curate 30 tracks spanning a wide emotional and genre range. Think social-first: what would someone actually put on their TikTok, Reel, or story? The selection should feel like a playlist a tastemaker friend made — not a stock music library. We want tracks that feel lived-in, current, and natural.
Key principle: This is social media music, not ad music. No corporate anthems, no generic upbeat ukulele, no "inspirational" piano. We want the stuff that organically goes viral — because it makes you feel something real.
Deliverable: 30 tracks — curated, tagged, and scored on our Oasis platform
Format: High-quality audio files (WAV or 320kbps MP3)
Sources: Royalty-free libraries, Creative Commons, direct licensing from independent artists, production music catalogs
Tagging Schema — 3-Axis Classification
Every track is classified using three axes that speak the language of video editors. All tagging and scoring is done on Oasis, our internal annotation platform. You may need a Feishu account for platform authorization — we'll confirm during onboarding.
Axis 1: Emotion — What does it make you feel?
The narrative intent of the music. Driven by mode (major/minor), harmonic tension (resolved vs unresolved), and melodic direction (rising vs falling).
| Category | Mood Tags |
|---|---|
| Uplifting | Happy, Uplifting, Carefree, Love, Ecstatic |
| Calm | Chill, Peaceful, Contemplative |
| Dark/Tense | Sad, Somber, Serious, Angry, Tense, Eerie |
Axis 2: Energy — How much weight does it carry?
Determines cut pacing and visual weight. Low energy supports slow edits; high energy demands fast cuts.
| Level | BPM Range | Character |
|---|---|---|
| Low | 0–84 | Soft, warm, sparse |
| Medium | 85–115 | Present, clean, balanced |
| High | 116+ | Loud, aggressive, dense |
Axis 3: Flavor — What world does it live in?
Defined by instrumentation, production style, and cultural associations.
| Category | Attributes |
|---|---|
| Rhythmic Urban | rap, punk, grungy, swagger, drums, bass |
| Song-Led/Intimate | singer-songwriter, acoustic guitar, piano, soft |
| Cinematic/Expansive | cinematic, ambient, strings, epic |
| Texural/Abstract | sci-fi, psychedelic, ethereal, glitchy, sound fx |
| Organic/Earthy | earthy, tribal, nature sounds, world |
| Playful/Stylized | whimsical, quirky, funky, fun |
Each track gets one tag from each axis — creating a 3-dimensional profile (e.g. Uplifting × Medium × Cinematic/Expansive). Some combinations will be common; others rare. Both are valuable for training.
04Music Categories
The 30 tracks should cover these broad categories, roughly weighted by how frequently LZ users will encounter these moments:
High Priority (~12 tracks)
- Daily life / casual moments
- Outdoor / travel / adventure
- Pets & family
- Content creation (vlogs, b-roll)
Medium Priority (~10 tracks)
- Cinematic / dramatic
- Night / urban / moody
- Sport / action / fitness
- Product / tech showcase
Lower Priority (~5 tracks)
- Romantic / intimate
- Nostalgic / retro
- Party / celebration
- Seasonal / holiday
Edge Cases (~3 tracks)
- Abstract / experimental
- Ambient / meditative
- Humorous / quirky
- Tension / suspense
05Deliverables & Format
| Deliverable | Format | Notes |
|---|---|---|
| 30 Tracks | Organized folder (by category → mood) | WAV or 320kbps MP3 |
| Oasis Scoring | Completed on platform | All 30 tracks tagged per 3-axis schema + any notes |
06Timeline & Compensation
This is a small, focused task — estimated at a few hours of work. Onboard to Oasis, review the schema, source and tag 30 tracks, done.
Compensation will be discussed during the interview stage. This is a short-term contract. Future work may follow depending on results.
07Ideal Candidate
- Deep knowledge of music across genres — not just one lane
- Experience in music supervision, editorial music selection, or playlist curation at scale
- Understanding of how music drives visual editing rhythm (video editors, music supervisors for film/TV, or content creators who edit to music)
- Meticulous and systematic — the tagging schema is critical and must be consistent
- Familiarity with music licensing (royalty-free, Creative Commons, sync licensing)
- Bonus: experience with AI/ML datasets, data annotation, or music information retrieval
08How to Apply
Please send:
- A brief intro and your availability
- Your hourly rate
- A sample playlist (10–15 tracks) that shows your range, taste, and diversity — we want to see that you understand what's working on social media right now and why certain songs organically take off
MDR — Confidential · March 2026