AI Development
Active engineering work at Quantum Factor AI — from autonomous agent frameworks and timer-driven systems to the PRISM middleware architecture and AIOS governance layer.
Timer-Driven Agent Framework
Two core agent programs under active development. The Timer Agent manages workload distribution across multiple tasks using dynamic scheduling and priority queuing. The Watchdog Engine provides the production reliability layer — detecting anomalies, triggering fault recovery, and ensuring continuous operation without human intervention.
Timer Agent
Polling-based workload balancing with real-time priority queue management. Adapts resource allocation dynamically as task dependencies change.
Watchdog Engine
Continuous monitoring of agent internal state. Detects deviations from baseline, triggers alerts or automated recovery, minimizes downtime.
Scheduling Model
Timer-driven polling with dependency graph evaluation. High-priority tasks preempt lower-priority queues at configurable intervals.
Fault Tolerance
Three-tier response: alert → fallback plan → automated recovery. Cascade failure prevention through isolated fault domains.
- Dynamic resource reallocation in response to changing task dependencies
- Priority queue data structure ensures critical tasks execute first
- Anomaly detection against predefined thresholds and KPI baselines
- Automated fault recovery with configurable fallback logic
- Designed for cloud-based and distributed system deployment
PRISM Middleware
PRISM is a logic-driven AI middleware designed to bridge high-fidelity data structures with autonomous execution layers. The architecture manages sovereign execution of complex logic sets, handles data routing between agent tiers, and provides the compliance and verification layer for production AI deployments.
- Five-ring architecture separating safety, governance, logic, agents, and roles
- Multi-provider AI routing — Anthropic, OpenAI, Gemini, Ollama
- Logic sharding for ISO 42001 and NIST AI RMF 2.0 compliance
- Sovereign local execution — air-gapped deployment support
AIOS Governance Layer
The AIOS Governance & Orchestration Layer is a single-file HTML/JS application backed by a FastAPI server. It implements a four-stage pipeline — ICO rewrite, constitution injection, governed response, drift audit — with real-time DDS scoring, vector-based Q-score computation, compliance attestation reporting, and a three-agent content engine.
Stage 1 — ICO Rewrite
Input reformulated into Intent / Constraints / Context / Output Standard. Memory-driven — adapts to user profile and project context.
Stage 2 — Constitution
14-section governance constitution (v18) injected into every request. Covers Ring architecture, failure modes, PRISM distillation, and PCVS.
Stage 3 — Response
Multi-provider streaming response. Supports Anthropic, OpenAI, Gemini, and local Ollama models. Provider selection persists across sessions.
Stage 4 — Audit
Vector Q-score via all-MiniLM-L6-v2 cosine similarity. DDS rolling score across 10 turns. Compliance attestation downloadable as .md.
- DDS visual display — 5-band color indicator (STABLE → UNSAFE) live in the UI
- Portable memory — export/import via JSON for thumb drive deployment
- Three-agent content pipeline — Synthesizer → Writer → Editor → .md download
- File upload to Content Engine — .txt, .md, .py, .docx, .pdf with structured manifest
- Template trigger shortcuts — TPL-001 through TPL-009 above the input bar
- NIST AI RMF 2.0, EU AI Act Art 14 & 50, Six Sigma compliance claims
Next Development Targets
Items queued for active development once current sprints complete.
- D10D3 Industrial Auditor — ISO 42001 compliance engine, currently paused
- AMBER Sovereign AI OS — local Phi-3.5 inference stack, currently paused
- Vector database integration (ChromaDB) for cross-session memory persistence
- Offline model deployment — Llama-3-8B via Ollama for air-gapped AIOS
- GitHub/Astro automated publishing pipeline from Content Engine output