Senior AI Product Engineer - Computer Vision & LLM Systems

Role Overview

Join Dermalytics as we complete development of our RGB-to-UV skin analysis platform and build AI-powered product recommendations. You'll take over an advanced training system (PyTorch Lightning + AWS SageMaker) with active experiments, complete model deployment, build inference services, and develop skin analysis features powered by computer vision and LLMs.

This is not a build-from-scratch role - you'll be completing in-progress work on a solid foundation and expanding into new features.

About Dermalytics

Dermalytics is building AI-powered skin analysis technology that translates smartphone RGB images into synthetic UV images for professional-grade skin assessment. Our platform helps people understand their skin health without expensive UV imaging devices, democratizing access to dermatological insights.

We're at an exciting inflection point:

• Professional ML training infrastructure built and running
• Active experiments on AWS SageMaker with real datasets
• Ready to deploy first production models
• Expanding into comprehensive skin analysis platform

What You'll Own

This role combines deep learning engineering (60-70%), full-stack AI development (20-25%), and internal automation (10-15%).

1. RGB→UV Model Completion (Months 1-2)

Take over active training experiments:

• Complete training runs for RGB→UV image translation on AWS SageMaker
• Validate model accuracy across diverse skin tones (Fitzpatrick I-VI) and lighting conditions
• Optimize hyperparameters for production quality (learning rate, features, epochs)
• Benchmark performance against MEICE device ground truth (SSIM, PSNR, MAE)
• Document model performance, limitations, and edge cases

What already exists:

• ✅ PyTorch Lightning training system (complete implementation)
• ✅ UNet architecture for image-to-image translation
• ✅ Experiments running with real MEICE dataset on S3
• ✅ Training metrics configured (MAE, SSIM, PSNR)
• ✅ AWS SageMaker infrastructure with GPU quotas

Success metrics:

• SSIM > 0.85 on validation set
• Model validated across all Fitzpatrick types
• Comprehensive benchmarking report
• Production-ready model weights

2. Production Inference Service (Months 3-4)

• Build FastAPI service from infrastructure:
• Develop inference API with model loading from S3/SageMaker
• Implement preprocessing pipeline for mobile RGB images
• Optimize model for mobile deployment (quantization, ONNX export)
• Deploy to AWS SageMaker endpoints with auto-scaling
• Add comprehensive monitoring, logging, and error handling
• Performance testing and optimization (<500ms latency target)

What already exists:

• ✅ AWS CDK stacks for SageMaker endpoints
• ✅ S3 buckets and IAM roles configured
• ✅ Model registry setup
• ❌ Inference service code (needs building)

Success metrics:

• API deployed to production with 99.9% uptime
• <500ms inference latency (p95)
• Model size reduced 50% for mobile
• Load testing passed (100+ concurrent requests)

3. Skin Analysis Platform (Months 5-7)

• Build comprehensive skin analysis features:
• Develop skin concern detection models (acne, dark spots, wrinkles, texture)
• Implement facial zone mapping for targeted analysis (T-zone, cheeks, forehead)
• Build LLM-powered recommendation engine using OpenAI/Anthropic APIs
• Create product database with ingredient analysis
• Design recommendation logic based on detected concerns
• Integrate all components into cohesive API
• Testing and validation across diverse use cases

What already exists:

• ✅ Face detection pipeline (4 methods: OpenCV, dlib, MediaPipe)
• ✅ Image preprocessing utilities
• ❌ Concern detection models (need building)
• ❌ LLM recommendation system (need building)

Success metrics:

• 5+ concern types detected with >80% accuracy
• LLM generating relevant, personalized recommendations
• Product database with 100+ SKUs
• End-to-end demo ready for user testing

4. Internal AI Automation (20% ongoing)

• Build custom AI bots for business operations:
• Design and implement 3-4 AI automation projects over 12 months
• Create LLM workflows using n8n + OpenAI/Anthropic APIs
• Integrate with business tools (ClickUp, Slack, Shopify, Gmail)
• Examples: Customer service bot, content generation, data analysis
• Maintain and optimize deployed automations

Success metrics:

• 3-4 bots deployed and actively used
• Measurable time/cost savings for business functions
• Positive feedback from internal users

5. LED Esthetics App Support (10% as needed)

• Maintain existing iOS/Android app:
• Bug fixes and performance improvements
• Feature implementation based on roadmap
• App store deployments and version management
• Coordinate with design team on UI/UX updates

Technical Stack

ML & AI:

• PyTorch, PyTorch Lightning
• torchvision, timm (model zoo)
• OpenAI API, Anthropic Claude API
• ONNX, TensorRT (optimization)
• scikit-learn, OpenCV

Cloud & Infrastructure:

• AWS SageMaker (training & inference)
• AWS S3, Lambda, API Gateway
• AWS CDK (Infrastructure as Code)
• Docker, GitHub Actions

##Backend:

• FastAPI, Pydantic
• SQLAlchemy, Alembic
• PostgreSQL or DynamoDB

Automation:

• n8n workflow automation
• LangChain (optional)
• OpenAI/Anthropic SDKs

Development Tools:

• Cursor (AI-assisted coding)
• Claude AI (code review, architecture)
• GitHub Copilot
• Pre-commit hooks (ruff, mypy, bandit)
• pytest

Qualifications

Required Skills

Deep Learning & Computer Vision:

• 3-5 years professional ML/AI engineering experience
• Strong PyTorch skills with production model deployment
• Understanding of encoder-decoder architectures (UNet, ResNet)
• Experience with image-to-image translation or similar CV tasks
• Knowledge of perceptual metrics (SSIM, PSNR, perceptual loss)
• Ability to evaluate and debug model performance

Production ML Engineering:

• Experience with PyTorch Lightning or similar frameworks
• Model optimization techniques (quantization, pruning, ONNX)
• Cloud ML platforms (AWS SageMaker strongly preferred)
• Model serving and inference optimization
• Understanding of distributed training concepts

Backend Development:

• Proficiency in FastAPI or Flask for API development
• RESTful API design and best practices
• Python best practices (type hints, testing, documentation)
• Async programming (asyncio, async/await)
• Error handling, logging, and monitoring

DevOps & Cloud:

• AWS services experience (SageMaker, S3, Lambda at minimum)
• Docker containerization
• CI/CD pipelines (GitHub Actions or similar)
• Infrastructure as Code familiarity (CDK, Terraform, or CloudFormation)
• Git workflow and collaborative development

AI Integration:

• LLM API usage (OpenAI, Anthropic, or similar)
• Prompt engineering and optimization
• Basic understanding of RAG (Retrieval Augmented Generation)
• API integration and error handling

Strongly Preferred

• PyTorch Lightning hands-on experience
• Image-to-image translation projects in portfolio
• AWS SageMaker training jobs experience
• Model deployment to mobile (CoreML, TensorFlow Lite, ONNX)
• Beauty, healthcare, or dermatology ML applications
• LangChain or similar LLM frameworks

Nice to Have

Computer vision research publications
GAN experience (Pix2Pix, CycleGAN) - helpful if UNet quality needs improvement
React Native, Swift, or Kotlin for app development
n8n workflow automation
Vector databases (Pinecone, Weaviate, Chroma)
Spanish fluency (for LATAM team collaboration)

Success Metrics by Timeline

Month 1-2: Model Training Complete

• RGB→UV model achieves SSIM > 0.85 on validation set
• Model validated across all Fitzpatrick skin types (I-VI)
• Hyperparameter optimization documented
  Benchmarking report completed vs. MEICE device
• Production-ready model weights in S3/SageMaker

Month 3-4: Inference API Deployed

• FastAPI service deployed to AWS SageMaker
• <500ms inference latency (p95) achieved
• Model optimized for mobile (50%+ size reduction)
• 99.9% uptime SLA maintained
• Load testing passed (100+ concurrent requests)
• Monitoring and alerting configured

Month 5-7: Skin Analysis Platform Live

• 5+ skin concerns detected with >80% accuracy
• Facial zone mapping implemented and tested
• LLM recommendation engine generating relevant suggestions
• Product database with 100+ SKUs and ingredient data
• End-to-end system demo completed
• User testing feedback collected

Month 6-12: Full Product & Automation

• RGB→UV model continuously improved based on user feedback
• 3-4 internal AI automation bots deployed and used daily
• LED Esthetics app maintained (4.5+ star rating)
• Platform handling production traffic reliably
• Documentation complete for all systems

Compensation & Benefits

Compensation:

• Contractor, paid monthly via Deel/Remote.com
• $1,500-2,000 equipment stipend (laptop, monitor)
• AI tools provided (Claude API, Cursor, GitHub Copilot)
• AWS credits for experimentation

Work Arrangement:

• Fully remote (LATAM, UTC-3 to UTC-6)
• Flexible hours, 4-hour core overlap (10am-2pm EST)
• Focus on results, not hours

Growth:

• Own complete ML pipeline from training to deployment
• Potential to build/lead AI team as company scales
• Direct impact on product direction

Why This Role is Compelling:

• Strong Foundation - Take over working PyTorch Lightning system with active SageMaker experiments, not starting from scratch
• Real Impact - Democratize professional skin analysis (normally $1000+ UV devices) for millions of users
• Technical Growth - Work with advanced CV (image-to-image translation), production ML at scale, and modern LLMs
• True Ownership - Own complete ML pipeline, make architectural decisions, minimal micromanagement
• AI-Accelerated - Leverage Claude/Cursor for rapid development and focus on high-level architecture
• Work-Life Balance - Remote-first, flexible hours, no commute, async-friendly culture

Ideal Candidate Profile

You're a great fit if you:

• Have built and deployed ML models to production
• Understand both the "research" and "engineering" sides of ML
• Can work independently with minimal direction
• Care about product quality and user experience
• Enjoy solving real-world problems with AI
• Communicate proactively in remote settings
• Balance perfectionism with pragmatism (ship working solutions)
• Use AI tools to accelerate your work

You're NOT a good fit if you:

• Only want to do pure research (this role ships product)
• Need constant oversight and direction
• Prefer working 9-5 in an office
• Only want to work on one narrow aspect of ML
• Don't enjoy cross-functional work (LLMs, backend, etc.)
• Dislike iterative development and user feedback