# ML Research Lead
## Wazza Studio LLC

**Location:** Remote (US/EU preferred) or Paris, France
**Type:** Full-time
**Compensation:** competitive + equity
**Reports to:** CEO (Co-Founder) and later on CTO

---

## About Wazza

Wazza is building the world's first IP-licensed AI model network for film and media production. We're creating production-grade video generation models trained on properly licensed content, with transparent blockchain-tracked royalty distribution to IP owners.

We just secured partnership with a major producer to train our first model and are building the infrastructure that will become the industry standard for compliant AI content generation.

**Our stack:** Diffusion Transformers (DiT), Pyramidal Flow Matching, 3D-aware generation, MegaETH/Solana blockchain, Next.js/NestJS, PostgreSQL.

---

## The Role

As our **first ML Research Lead**, you'll architect and train the foundational models that power our IP-licensed AI network. This isn't incremental research—you're building production systems that will generate film-quality video for Netflix distribution.

You'll work directly with the founding team to translate cutting-edge research (Sora, Movie Gen, Pyramidal Flow) into production models, starting with our producer's IP as the first proof of concept.

**This role is perfect for someone who:**
- Wants to build production AI systems, not just publish papers
- Thrives in 0→1 environments with high autonomy
- Cares about the legal/ethical implications of AI training
- Has strong opinions on model architecture (and can back them up)

---

## What You'll Do

### **Months 1-3: Foundation**
- Design training architecture for first DiT model (3B-7B params)
- Set up distributed training infrastructure on AWS (8-16 A100s)
- Build data preprocessing pipeline for producer's IP (50+ hours of footage)
- Establish evaluation metrics (FVD, CLIPSIM, face similarity)
- Train baseline text-to-video model

### **Months 4-6: Camera Control & Identity**
- Integrate 3D cache system (GEN3C-style) for precise camera control
- Implement frequency-domain identity preservation (ConsisID approach)
- Optimize inference speed (<15s for 512p, <30s for 1080p)
- Validate with pilot users, iterate based on feedback

### **Months 7-12: Scale & Productionize**
- Train 2-3 additional IP models with optimized pipeline
- Implement hybrid model generation (multi-IP style blending)
- Collaborate with Abraxas team on automated royalty attribution
- Scale inference infrastructure (self-hosted K8s deployment)
- Mentor growing training engineering team (3-5 reports)

### **Ongoing Responsibilities**
- Stay current with latest research (CVPR, ICLR, NeurIPS)
- Evaluate new techniques for production viability
- Represent Wazza at conferences and with research partners
- Build relationships with academic labs for collaboration
- Patent filings for novel architectural innovations

---

## Requirements

### **Must Have**
- **PhD in ML/CS** or equivalent research experience (5+ years industry)
- **Hands-on experience training large models** (1B+ parameters)
- **Deep understanding of diffusion models** and/or transformers
- **Production ML experience** (not just research prototypes)
- **Strong Python/PyTorch** skills, comfortable with distributed training
- **Track record of shipping**: models that actually run in production
- **Communication skills**: explain complex concepts to non-ML stakeholders

### **Ideally Have**
- Experience with **video generation models** (Sora, Runway, Pika, etc.)
- Published at top-tier ML conferences (CVPR, ICLR, NeurIPS)
- Worked with **large-scale GPU clusters** (100+ GPUs)
- Experience with **3D vision** (depth estimation, camera pose, NeRF/Gaussian Splatting)
- Understanding of **film production workflows** and quality requirements
- Entrepreneurial mindset: built things from scratch, comfortable with ambiguity

### **Nice to Have**
- Interest in blockchain/web3 (we use Solana for royalty tracking)
- Experience with **model optimization** (distillation, quantization, pruning)
- Familiarity with **MLOps tools** (Weights & Biases, MLflow, Kubeflow)
- Creative background (filmmaking, photography, visual arts)

---

## What Success Looks Like

### **Month 3**
- ✅ First model trained: 512p text-to-video working
- ✅ FVD <200, CLIPSIM >0.28 on benchmark
- ✅ Training infrastructure documented and reproducible

### **Month 6**
- ✅ Camera control operational (3D-aware generation)
- ✅ Identity preservation: face similarity >0.90 across shots
- ✅ Pilot users generating content successfully
- ✅ Technical roadmap for next 3 models defined

### **Month 12**
- ✅ 3 production models deployed and generating revenue
- ✅ Training cost optimized by 30-50% vs initial model
- ✅ Team of 3-5 training engineers hired and ramped
- ✅ Architecture contributing to Wazza's competitive moat

---

## Why Join Wazza?

### **Impact**
- Build foundational AI models for a $50B+ industry transformation
- Your work will directly enable independent filmmakers to compete with studios
- Legal, compliant approach means your models will actually be used (not banned)

### **Technical Challenge**
- Cutting-edge research: DiT, flow matching, 3D-aware generation, identity preservation
- Scale: Training multi-billion parameter models on clusters of A100s
- Production constraints: Film-quality output, <30s inference, Netflix-compliant

### **Autonomy**
- Design the architecture, choose the frameworks, set the research direction
- Direct access to founders (no bureaucracy)
- Budget to experiment and fail fast

### **Upside**
- Early employee (employee #6-10) with meaningful equity
- If we become the "Spotify of AI model licensing," your stake is life-changing

### **Team**
- Work with co-founder who's shipped production AI systems (Wazza Engine: 102 models)
- Cross-functional: ML, blockchain, film production, legal—all under one roof
- Remote-first culture with in-person offsites

### **Growth**
- This role grows into VP of AI Research as we scale
- Build and lead a world-class research team (10-20 people)
- Shape the future of AI in media production

---

## Compensation & Benefits

**Cash:** Competitive (based on experience and location)
**Equity:** Potential (4-year vest, 1-year cliff)
**Benefits:**

- Unlimited PTO (minimum 3 weeks required)
- Annual learning budget (courses, conferences, books)
- Latest hardware: MacBook Pro M3 Max + NVIDIA GPU workstation
- Relocation assistance available (if moving to Paris)

**Compute Budget:** Whatever you need. Seriously. We have AWS credits and GPU reservations ready.

---

## Interview Process

1. **Initial Screen (30 min):** COO - fit, motivation, logistics
2. **Technical Deep-Dive (90 min):** CEO - architecture design, past projects
3. **Research Discussion (60 min):** Discuss recent papers, technical approach for our use case
4. **Founder Chat (45 min):** Vision alignment, company strategy, equity discussion
5. **Reference Checks**
6. **Offer**

**Timeline:** 2-3 weeks from application to offer

---

## How to Apply

**Email:** [e-mail]
**Subject:** "ML Research Lead - [Your Name]"

**Include:**
1. **Resume/CV** (PDF)
2. **Brief cover letter** (max 500 words) answering:
- Why are you interested in AI for film production?
- What's your take on the technical approach outlined in our architecture? (We can share details)
- What's a model you've trained that you're proud of? What made it challenging?
3. **Links:** GitHub, Google Scholar, personal site (optional but appreciated)

**Bonus points if you:**
- Send a video introduction instead of cover letter (we're a video company after all!)
- Share thoughts on Sora, Movie Gen, or recent video generation research
- Include examples of production systems you've built

---

## Our Values

🎯 **Ship First, Publish Later** - Production systems over papers
⚖️ **Legal > Cool** - Compliance and ethics are non-negotiable
🚀 **Move Fast, Think Deep** - Speed + rigor, not speed vs. rigor
🤝 **Create Value for Creators** - IP owners should profit from AI
🌍 **Remote Default** - Best talent is global

---

**Questions?** Email us at [e-mail]

**Learn more:** wazza.ai

---

*Wazza Studio LLC is a Delaware LLC building the future of AI-powered media production.*