AI & ML interests
Piqued AI builds governance-aware AI for collaborative, regulated environments. Open standards: the pqdx universal state protocol and pqdc software-development domain profile. Active work spans open-weight model fine-tuning, MCP-native agentic architecture, and memory-aware AI substrates. Local-first by default.
Piqued AI
Governance-aware AI for collaborative, regulated environments.
What we build
- Open standards — pqdx (universal state protocol) and pqdc (software-development domain profile).
- Two model families — base substrates (SymbiOS-Core) and curation engines (Synapse-Engine), published on Hugging Face for customers to fine-tune privately on their own corpus.
- Local-first deployment — designed for on-prem, edge, and air-gapped environments. Your data stays on your substrate.
Models
Two parallel families, sized for community fine-tuning and on-prem inference.
SymbiOS-Core — memory-aware base substrate
Foundation models with the scaffolding for memory-awareness baked in. Published as starter substrates — generic, org-naive, ready to fine-tune. Each customer's operational substrate is a private fine-tune produced by running Synapse-Engine over their own corpus; that private model stays on the customer's substrate and never publishes.
Sizes: 1.5B · 7B · 14B · 32B
Synapse-Engine — curation and orchestration
Instruct-tuned, MCP-native models that take raw corpora (conversations, decisions, governance artifacts) and turn them into the curated training data that produces a SymbiOS-Core fine-tune. The methodology made executable.
Sizes: 1.5B · 7B · 14B · 32B
All repos currently private — namespace reservation phase. Weights, model cards, and licenses publish per-tier as each model ships.
Contact
- Piqued AI — piqued-ai.com
- Parent (holding) — Our Collaborative Space
- Sister entity — Piqued Solutions — cybersecurity, DevSecOps, AI advisory
