Guardian of Sorrows and Joys (Machine Learning Engineer, Small Language Models & Privacy)

Company description

Machines of Grace is building Theotokos AI, an interactive prayer journal. It meets people wherever they are, from grief and distress to the quiet daily work of growth and self-awareness. Each entry becomes an Intercession: the person’s words gathered into prayer, with a reflection, a meditation and gentle next steps. Privacy is the heart of the product. We are a small founding team, so this hire shapes the product and its culture from the start.

Role description

This is a full-time, remote role in Canada. Every day, people write their most private sorrows, prayers, and dreams into Theotokos AI. Your mission is to build a small language model, and the privacy architecture around it, so that what users write stays protected. Ideally, it is processed on their own device or somewhere no one, including us, can read it.

You will

- Design, fine-tune and deploy a small language model (Llama, Mistral, Phi or Gemma class) that turns journal entries into intercessions, using LoRA, distillation and quantization
- Run inference on-device or in secure, isolated environments
- Architect end-to-end and client-side encryption for journal entries, including key management, rotation and secure deletion




- Build a safety layer that detects distress and routes users to real help immediately
- Create evaluation suites for quality, tone, safety and privacy leakage, and red-team each release
- Lead privacy-by-design work for Quebec Law 25 and PIPEDA

- Qualifications

- 5+ years in machine learning, with hands-on fine-tuning and deployment of open-weight language models (PyTorch or similar)
- Strong applied cryptography and security engineering: encryption at rest and in transit, key management, threat modelling
- Experience with on-device or edge inference (llama.cpp, ONNX, MLX, WebGPU or similar)
- Python plus TypeScript or another production language; comfort with Next.js and PostgreSQL
- Sound judgment on trade-offs between model quality, speed, cost and privacy
- Transparent communication with non-technical collaborators
- Assets: differential privacy, federated learning or confidential computing; experience with journaling, health or other sensitive-data products; working French; a degree in CS, math or a related field, or equivalent experience
- An interest in contemplative technology and supporting people through every season of life

Salary: $110,000 - $140,000 CAD · [equity]

📌 Machine Learning Engineer — Small Language Models & Privacy (Canada)
🏢 Machines Of Grace
📍 Canada