Inspiration
As a student working on my graduation thesis, I constantly struggled with turning vague research ideas into structured plans, literature reviews, and academic figures. Existing AI tools are powerful at generating text, but they often lack structure, consistency, and the ability to maintain long-term research context. I wanted an agent that could truly act as a research collaborator — not just a chatbot.
Kiroween’s theme of “bringing something dead back to life” inspired me to resurrect the traditional research workflow with modern agentic capabilities.
What it does
GhostAgent is an agentic research assistant built with Kiro. It helps students and researchers:
- Turn a rough thesis idea into structured Specs (requirements, design, and task breakdown)
- Automatically generate literature review summaries and research outlines
- Create academic diagrams and architecture figures (using tools like Mermaid / TikZ style descriptions)
- Maintain long-term research context across multiple sessions through Kiro’s steering and memory capabilities
The goal is to make the early and middle stages of thesis writing significantly more efficient and less chaotic.
How we built it
GhostAgent is built entirely with Kiro, leveraging its core strengths:
- Spec-driven development: The entire system was designed using Kiro Specs (requirements → design → tasks)
- Multi-agent workflow: Different specialized agents handle literature search, structure planning, writing assistance, and diagram generation
- Steering documents & Hooks: Used to enforce academic writing standards and consistent output format
- MCP integrations: Connected external tools for literature and diagram generation
The project structure and agent behavior are defined inside the .kiro directory, following Kiro best practices.
Challenges we ran into
- Keeping the agent’s output academically rigorous while still being creative
- Designing good steering rules so different agents don’t conflict with each other
- Balancing structure (Specs) with flexibility (vibe coding for quick experiments)
- Making the diagram generation actually useful for real thesis figures
Accomplishments that we're proud of
- Successfully turned a complete research idea into a structured Spec using Kiro
- Built a working multi-agent research pipeline
- Created a system that feels like a real research collaborator rather than just another AI chatbot
What we learned
- Spec-driven development is extremely powerful for complex agent systems
- Good steering documents are more important than complex prompts
- Kiro’s multi-surface capability (IDE + CLI) makes iterative development much smoother
What's next for GhostAgent
- Add better literature management and citation handling
- Support more academic figure types
- Improve long-term memory for multi-month thesis projects
- Explore team collaboration features for research groups
Built With
- academic
- agent
- ai
- assistant
- development
- ide
- kiro
- literature
- mermaid
- multi-agent
- python
- research
- review
- spec-driven
- steering
- thesis
