Mistakes and errors are inevitable in enterprise systems. Developers introduce bugs, product teams make imperfect decisions, and users sometimes perform unexpected actions that cause data inconsistencies. Once such errors leave development and reach staging or production, the cost of mitigation grows — requiring coordination between multiple teams and risking business disruption.
For my Master’s thesis in Business Informatics at Riga Technical University, I explored innovative methods to address functionality gaps, data integrity violations, and system interoperability issues resulting from these types of errors.
I conducted a systematic literature review to map the current research landscape and then proposed a method that integrates an AI agent to assist in mitigating errors quickly and cost-effectively.
👉 GitHub – Healer Practical Experiment: matissg/healer
What Does the Healer Practical Experiment Do?
The experiment demonstrates how error information and source code artefacts can be passed as a prompt to an AI agent, which suggests temporary fixes or adaptations to unblock users. Rather than modifying data permanently, the system aims to adapt behavior based on known error patterns.
Key Features:
- ⚙️ Detect tactical-level application errors
- 🤖 Sends prompts with contextual error info to an AI agent
- 🔄 Receives and applies adaptive code
Findings and Reflections
- ✨ The approach showed promise in resolving common runtime failures quickly.
- ⚠️ AI agent responses varied in quality — same prompts could return different outcomes.
- 🧪 Fast feedback loops and minimal intervention requirements were key benefits.
- 💡 The method’s low cost and speed make it appealing for non-critical or interim solutions.
Despite its limitations, the method offers a low-overhead, AI-assisted way to reduce the impact of system errors on end users.
My thesis contributes to the growing body of work on AI-assisted error handling and opens a path toward more adaptive, resilient enterprise software systems.