Elik Plim Kudowor

Project Mirror — Multi-Agent AI System for Workflow Automation

Project Mirror is a multi-agent AI system designed to act as a professional assistant, automating tasks such as information retrieval, scheduling, and technical reasoning through coordinated agent workflows.

The system is built to demonstrate how AI can move beyond isolated capabilities into reliable, task-oriented tools used in real workflows, where outputs directly influence decisions and actions.

Key Contributions:

• Designed a hierarchical multi-agent architecture for task decomposition and execution
• Implemented retrieval-augmented reasoning (RAG) with strict context isolation to improve accuracy
• Integrated external tools (e.g., scheduling APIs) for real-world task execution
• Improved system reliability through structured outputs and multi-layer validation
• Built observability mechanisms to track system performance and debug agent behavior

Impact:

• Reduced complex workflow execution time from hours to minutes
• Significantly improved output reliability through validation and adversarial testing
• Demonstrated a scalable approach to building production-oriented AI systems

Tech: Python, FastAPI, Google ADK, MCP, Weaviate, LLM APIs (Gemini, Llama), Docker

Developed Kognia AI, a sophisticated multi-agent system designed to eliminate the “Cognitive Bottleneck” in brand strategy. Kognia moves beyond simple chatbots by deploying a specialized team of autonomous agents that collaborate to transform raw market data into high-stakes strategic insights.

Key Technical Highlights:

• Architecture: Implemented a Hierarchical Orchestration Pattern using the Google Agent Development Kit (ADK) to prevent context window overload and minimize hallucinations.
• Orchestration: Engineered the kognia_nexus_agent to act as a central project manager, dynamically routing tasks to specialist agents based on user intent.
• Grounding & Accuracy: Integrated Google Search Grounding and deep scraping tools to ensure all strategic outputs are based on real-time market facts rather than static training data.
• Synthetic Focus Groups: Developed a Conversation Simulator Agent that allows users to “interview” data by roleplaying as specific consumer personas, providing a novel way to stress-test brand messaging.

The Specialist Agent Team:

• Market Intel Analyst: Automates deep-web research and trend discovery.
• SWOT Evaluator: Applies rigid SWOT frameworks to identify opportunities and threats.
• Report Architect & Executive Briefer: Synthesizes complex multi-agent outputs into client-ready documents and high-impact leadership summaries.

Impact:

• Reduced time taking for manual research and strategic insight generation from days to < 5 minutes.

Tech Stack:
Python | Google Gemini | Google ADK | Google Search Grounding | Generative AI | Multi-Agent Systems (MAS)