Multi-agent content generation platform

VirtuAI

A capstone platform that orchestrates specialized agents and turns one knowledge base into coordinated text, image, video and publishing workflows.

Period
April to June 2026
Team
3-person capstone team, Bahcesehir University
Role
Service architecture and agent orchestration
Technologies
  • CrewAI
  • LangGraph
  • PostgreSQL
  • REST APIs
  • OAuth 2.0

Key contributions

  • Designed the service architecture and implemented agent orchestration with CrewAI and LangGraph.
  • Built a PostgreSQL-backed knowledge base and a multimodal pipeline producing text, image and video output.
  • Built automated publishing workflows for five platforms and handled OAuth 2.0, token rotation and rate limits.

Problem

Content production across several formats and destinations creates coordination work: research has to stay consistent, outputs need specialist treatment and publishing integrations must remain reliable.

Context and responsibility

VirtuAI was Ayhan’s capstone project at Bahcesehir University, delivered by a three-person team. Ayhan designed the service architecture and implemented the agent-orchestration layer.

Architecture

The system connected CrewAI and LangGraph orchestration to a PostgreSQL-backed knowledge base. A multimodal pipeline produced text, image and video outputs before platform-specific publishing workflows handled delivery.

Important decisions

  • Specialized orchestration was separated from the shared knowledge layer.
  • Platform publishing used REST APIs rather than browser automation.
  • OAuth 2.0, token rotation and rate limits were treated as part of the integration design.

AI and context strategy

The verified source establishes a multi-agent architecture and a shared knowledge base. Exact model choices, prompt designs, retrieval metrics and token-cost figures are not publicly documented.

Security and reliability

OAuth token lifecycle and provider rate limits were explicit constraints. Additional security controls and production observability details are not publicly documented.

Outcome

The team delivered an end-to-end platform that could coordinate multimodal generation and automate publishing across five platforms.

Back to projects