AI is sandboxed from control
Research agents propose hypotheses, experiments, and analyses. The risk gateway, approval policies, and order execution are deterministic services that AI cannot call directly.
Architecture
Evidence-to-execution infrastructure.
QEV is a layered system: research and validation services on top, deterministic execution at the bottom, and an audit trail through everything. This page describes the target architecture; the roadmap states what is built at each stage.
EX-01 · System map
Five layers, one record.
Requests flow down through validation and control; evidence flows back up into memory.
Data and control flow
IntelligenceValidationQuantExecutionMemory
Frontend
Marketing website
Research workspace
Experiment dashboard
Operations console
Application API
Authentication
Projects
Strategy registry
Experiments
Deployments
Audit logs
Research services
AI orchestration
Document / RAG retrieval
Backtesting adapters
Statistical validation
Report generation
Market services
Market-data ingestion
Time-series storage
Symbol normalization
Corporate actions
Alternative datasets
Execution services
Paper simulator
Risk gateway— deterministic, outside the AI layer
Broker-neutral order schema
Tradier adapter— first adapter target
Future broker adapters
Infrastructure
Postgres
Object storage
Time-series database
Queue / workflow engine
Containers
Monitoring
EX-02 · Design principles
The rules the system is built around.
Everything is a record
Datasets, experiments, assumptions, code versions, approvals, and orders are written to the same audit trail. A live position can be traced back to the hypothesis that produced it.
Brokers are adapters, not foundations
Orders are expressed in a broker-neutral schema. Tradier is the first adapter target; Interactive Brokers and others follow the same interface. QEV integrates with quant engines such as LEAN — it sits above them.
Stages are enforced, not suggested
A strategy cannot reach live capital without passing paper simulation and explicit human approval. The gates are part of the data model, not a checklist in a document.
EX-03 · Stack direction
Boring where it should be, sharp where it matters.
Next.js and TypeScript on the surface; Postgres, time-series storage, and a workflow engine underneath; Python research workers where the math happens; containers for deployment. LEAN remains an optional quant engine — never a dependency of the brand.
web Next.js · TypeScript · Tailwind
api TypeScript or FastAPI services
data Postgres · object storage · time-series DB
workflows queue/workflow engine · Python workers
quant LEAN (optional) · custom Python
execution broker-neutral schema · Tradier first
deploy containers · Kubernetes later, not first
This stack describes the direction of the application platform. The marketing site you are reading is intentionally simpler — see the repository README for what powers this page.