Zyphra Tech Stack

Full-stack AGI company building open-source models and agentic systems

Technology, Information and Internet San Francisco, California 51–200 employees Privately Held

Zyphra is a full-stack AGI company operating a heavy research and engineering footprint (13 of 15 active roles), with concentrated hiring in research and core ML engineering. The tech stack reflects deep infrastructure work: PyTorch, vLLM, Ray, and SGLang for model training and inference; Kubernetes and Slurm for orchestration at scale; and active adoption of RAG plus browser automation (Puppeteer, Playwright) suggesting expansion into retrieval-augmented and agentic pipelines. Current project focus spans large-scale audio model training, novel architectures, and secure agent runtimes—indicating a shift from pure language models toward multimodal and interactive systems.

Tech Stack 43 technologies

Core StackPyTorch Python Ansible Terraform AWS Docker Kubernetes JavaScript TypeScript React Figma Playwright RAG Apache Spark Azure GCP Apptainer Slurm vLLM Ray SGLang Windows macOS HTML CSS Electron Tauri Puppeteer Apache Beam FAISS+13 more

AdoptingRAG Playwright Puppeteer

What Zyphra Is Building

◆Challenges

Performance optimization of training stack
Improving core modeling capabilities
Solving fundamental bottlenecks in contemporary models
Reliability of ml workloads
Scalability of compute environments
Incident response improvement
Data gathering and processing
Designing efficient architectures for gpu hardware
Compliance and access control in data handling
Building active community

▲Active Projects

Large-scale audio training runs
Novel model architectures
Search and retrieval pipelines across large-scale structured and unstructured data
Next generation open-source text-to-speech and audio models
Kernel development and optimization for large-scale ml workloads
Build and deployment systems
Release processes
Agentic systems and interaction projects
Secure virtualized runtimes and backend services for agent execution
Observability systems

Hiring Activity

Decelerating15 roles · 3 in 30d

Department

Engineering

Research

Data

Seniority

Mid

Lead

Senior

Company intelligence

Find more companies like Zyphra by tech stack, pain points and active projects

Get started free

About Zyphra

Zyphra is a San Francisco-based AGI company with 51–200 employees developing open-source models and infrastructure for large-scale machine learning. The company operates across three core areas: model research and training (audio and novel architectures), search and retrieval systems for structured and unstructured data at scale, and agentic systems with secure execution runtimes. Their engineering roadmap includes kernel optimization for GPU workloads, observability and reliability improvements for ML pipelines, and deployment tooling (Terraform, Ansible, Docker, Kubernetes) to support distributed training runs. The organization is U.S.-based and actively hiring in research and engineering roles.

HeadquartersSan Francisco, California

Company Size51–200 employees

Hiring MarketsUnited States

Frequently Asked Questions

What is Zyphra's tech stack?

PyTorch, Python, vLLM, Ray, SGLang, Kubernetes, Slurm for model training and inference. AWS, Azure, GCP for cloud. React, Electron, Tauri for frontend. Adopting RAG, Puppeteer, and Playwright for retrieval and automation.

What is Zyphra working on?

Large-scale audio model training, novel model architectures, search/retrieval pipelines, open-source text-to-speech, GPU kernel optimization, agentic systems with secure runtimes, and observability infrastructure for ML workloads.