Inspiration

Mainframes still run the core of banking, telecom, insurance, and retail. But real-time visibility into what’s happening inside them is extremely limited. Operators rely on batch SMF/JES logs, manual analysis, and decades of domain expertise to understand failures like ABENDs.

We wanted to show how modern cloud technologies — Confluent Cloud and Google Gemini — can bring mainframe operations into the real-time, AI-driven era. The idea: Stream mainframe-like events live through Confluent Analyze errors and ABENDs instantly with Gemini Visualize everything in a clean real-time dashboard

This bridges the gap between legacy infrastructure and modern AI.

What We Built

MainframePulse RT is a real-time mainframe intelligence system that turns operational events into instant insights.

Key features:

Live event streaming using Confluent Cloud (Kafka + SASL/SSL)

AI-powered diagnostics using Gemini 1.5 Flash

ABEND analysis

Root cause explanation

Recommended fix

Severity scoring

Confidence levels

Category classification

Real-time dashboard showing

Event feed

Severity breakdown

ABEND frequencies

Auto-scrolling logs

Click-to-expand detailed analysis panel

Start/Pause stream controls

Live indicator

Synthetic workload generator

Generates realistic SMF30, JES2, CICS, DB2, I/O, CPU spike events

15+ ABEND codes

Safely throttled to 5 events per 30 seconds

Everything runs live end-to-end.

How We Built It

Backend

Node.js + Express

KafkaJS consumer and producer

Confluent Cloud with SASL_SSL authentication

Gemini 2.5 Flash via the official Google Generative AI SDK

Custom event processor with in-memory ring buffer

Frontend (Dashboard)

Vanilla HTML/CSS/JS (no frameworks)

Auto-poll every 2 seconds

Live charts using

Animated table updates

Right-side event detail drawer

Real-time log panel

Data Pipeline Synthetic Events → Kafka Producer → Confluent Cloud → Kafka Consumer → Gemini Analysis → Event Processor → Real-Time Dashboard

What We Learned

How to set up secure Kafka streaming with Confluent Cloud (SASL_SSL)

Handling TLS issues and broker rotation with KafkaJS

Managing rate-controlled synthetic event streams

Using Google Gemini effectively for structured technical reasoning

Building a real-time dashboard without any frontend frameworks

Designing a full data pipeline from ingestion → AI → visualization

We also learned how powerful AI becomes when combined with streaming data — mainframe ops finally feel modern.

Challenges We Faced

TLS handshake failures when connecting KafkaJS to Confluent

Model migration issues — older Gemini models were deprecated

Producer/Consumer config mismatch (different config keys)

Stream rate control to avoid overloading Confluent or Google APIs

Designing a clean real-time UI with vanilla JS

Each issue taught us something new about building reliable streaming systems.

What’s Next

Integrate real SMF/JES logs

Add anomaly detection + predictive models

Add alerting & escalation workflows

Build multi-topic support for DB2, IMS, CICS, MQ streams

Package as a full SaaS for enterprise mainframe operators

Built With

Share this project:

Updates