TrustVote AI

Appearance

Agentic Workflow & Orchestrator Model ​

Context ​

As TrustVote AI scaled in complexity, the need for a robust, auditable, and production-safe delivery process became critical. Traditional single-agent or ad-hoc workflows led to bottlenecks and inconsistent quality. To address this, we adopted an agentic workflow centered on a dedicated Orchestrator agent.

Decision ​

Adopt a multi-agent workflow where a central Orchestrator agent owns end-to-end feature delivery, delegating work to specialist agents for each domain:

  • Backend Engineer (NestJS/Fastify, API, services)
  • Frontend Engineer (Next.js, UI, dashboard)
  • Data and AI Engineer (schema, migrations, embeddings, RAG)
  • QA and Quality Engineer (testing, coverage, regression)
  • DevSecOps Engineer (CI/CD, Docker, dependency security)
  • Security Engineer (threat modeling, cryptography, auth hardening)
  • Docs and ADR Engineer (docs, ADRs, engineering logs)
  • Research Engineer (library evaluation, RAG strategy, technical research)

The Orchestrator triages requests, routes work, and integrates all changes, enforcing quality, security, and documentation standards before completion.

Rationale ​

  • Separation of Concerns: Each agent brings deep expertise to its domain, reducing errors and increasing delivery speed.
  • End-to-End Ownership: The Orchestrator ensures no step is skipped, integrating outputs and running validation before completion.
  • Auditability: Every change is traceable, with logs and ADRs updated as part of the workflow.
  • Quality & Security: Automated enforcement of lint, test, and SonarCloud gates, plus built-in security review.

Consequences ​

  • Consistent, production-safe delivery across all packages.
  • Clear handoff and review sequence for multi-domain changes.
  • Documentation and logs are always updated in sync with code.
  • Increased transparency for all engineering decisions and changes.

This model is now the foundation for all engineering work in TrustVote AI.