Event Dates: June 26 – 29, 2025
Team Name: Dev Dream
Mainframe performance issues can cost Fortune 500 companies up to $2.5 million per hour. Solving them today requires deep technical expertise in legacy systems like SMF, RMF, and MXG.
Our solution: a voice-driven, multilingual, AI-powered assistant that allows users — even non-technical executives — to simply ask, "Why are payments slow?" and receive:
- Real-time diagnosis across subsystems like CICS, MQ, DB2, WAS
- Clear, actionable fixes
- Automated execution when needed
No more deciphering cryptic logs. Just ask and act.
Watch the Demo video: https://youtu.be/qZzBTY5bomk
"We built a tool where anyone — even your manager — can talk to a mainframe like talking to a person."
- Speak or type in natural language (in 7+ languages)
- Get answers instantly: what's wrong, where, and why
- Let the system fix issues automatically, if approved
All this is done using cutting-edge IBM Granite LLMs, a multi-agent architecture, and secure mainframe integration.
| Component | Technology Used |
|---|---|
| LLMs | IBM Granite 3.3 8B Instruct, Granite 3.2B |
| Architecture | Multi-Agent (Speech, Analysis, Edge, Retrieval) |
| Integration | ZOAU, USS Python, REST APIs |
| AI Pattern | RAG (Retrieval-Augmented Generation) |
- Speech Agent: Converts voice to prompt (simulated transcription)
- Analysis Agent: Queries mainframe logs, explains causes
- Edge Agent: Fast replies for basic status
- Retrieval Agent: Pulls relevant data in real-time using RAG
| Problem | Our Solution |
|---|---|
| $2.5M/hour outage costs | Instant diagnosis via voice |
| Hard-to-use legacy tools | No training needed — just speak |
| Lack of expert availability | AI fills in knowledge gaps automatically |
- Built in 62 hours during IBM TechXchange Hackathon
- Fully functional voice + mainframe interface using IBM LLMs
- Shows the future of mainframe automation
We loved working with IBM’s cutting-edge tech, especially the Granite family of models. This project opened doors to making legacy systems speak our language — literally.