Building secure, private, and trustworthy infrastructure for AI and data analytics.
Spire Studio creates research-grade open-source systems for privacy-preserving AI assistants, federated learning orchestration, adversarial evaluation, and cryptographic tooling. Our work sits at the intersection of AI security, privacy-preserving computation, and reproducible data workflows.
- Privacy-preserving AI assistants: local-first privacy layers that keep sensitive data out of remote model calls.
- Federated learning orchestration: reproducible experiment runners for distributed and privacy-aware ML.
- Attack and defense evaluation: standardized benchmark arenas for trustworthy AI and FL security research.
- Cryptographic tools for AI: practical libraries for privacy-preserving analytics and trustworthy computation.
| Project | What it does | Stack |
|---|---|---|
| CloakBot | A privacy-first AI assistant that sanitizes prompts locally before forwarding them to remote LLM APIs. | Python, local LLMs, LiteLLM |
| Figaro | An intelligent federated learning platform that turns natural-language experiment requests into reproducible FL runs. | FastAPI, LangGraph, PyTorch, React |
| FedArena | A standardized attack/defense evaluation arena for federated learning security research. | FastAPI, PyTorch, React, SQLite |
| PyFE4AI | A research-oriented Python library for functional encryption in trustworthy AI systems. | Python, cryptography |
- Local-first privacy: sensitive user data should stay on the user's machine whenever possible.
- Reproducibility first: experiments should be configured, tracked, replayable, and comparable.
- Standardized evaluation: trustworthy systems need shared benchmarks, not only one-off demos.
- Open infrastructure: research prototypes should be inspectable, extensible, and useful to builders.
- Use CloakBot to add a local privacy boundary in front of AI assistants.
- Try Figaro to run federated learning experiments from natural-language descriptions.
- Submit methods to FedArena and compare them against a benchmark matrix.
For collaboration or research inquiries, contact the Spire Studio maintainers through GitHub.