Case Study: Keysight Technologies Agentic AI Assistant
Nextbrick partnered with Keysight Technologies to build a transformative Agentic AI Assistant that replaces traditional search with a conversational, multi-turn reasoning engine.
The Challenge
Keysight needed to migrate from a legacy Coveo implementation to an Elastic-backed infrastructure while significantly improving the customer support experience. Users struggled to find precise technical information across a massive corpus of product manuals, service orders, and knowledge base articles.
The Solution
Nextbrick deployed the Nextbrick Agentic AI Platform to orchestrate a sophisticated RAG architecture, integrating natively with Salesforce while scaling the solution to index 30,000+ web pages across 8 languages.
- Salesforce Connector
- Multilingual RAG Pipeline
- Intelligent Routing & Guardrails
- Low-code Agentic Workflows
Multi-turn Reasoning & Result Execution
Contextual Routing & Local LLMs
The AI Assistant intelligently categorizes queries to serve fast-path precomputed answers (FAQ) or executes deep RAG validation loops over support docs utilizing localized, rapid LLMs like Qwen3 Coder.
Real-time Observability & Search Analytics
APM — Services Dashboard
Nextbrick provides live performance monitoring for all Keysight AI services, tracking latency, throughput, CPU, and error rates with real-time refreshes.
Search Performance Dashboards
Comprehensive analytics for search behavior, including top queries, zero-result patterns, and click-through rates across all support channels.
Comprehensive System Requirements
Nextbrick delivered against a complex matrix of operational, linguistic, and analytical capabilities.
Advanced Multi-Turn Context
- Q: "How do I make an eye diagram using ADS?"
- Q: "What is jitter?"
- Q: "Are there any products with max sample rate 2 GSa/s?"
- Q: "Find the instructions manual of U1610A product."
- Q: "How many MHz bandwidth is the U1610A product?"
Account & Operations Integration
- Task: "Where is my order? (4047199)"
- Task: "Give me case status? (00001036)"
- Task: "Create Support Case using subject + description + product details"
- Task: Pricing Lookups querying dynamic data from HTML, Elasticsearch, and PIM platforms.
Complex Translation & Guardrail Logic
Strict Native Language Fulfillment
When content exists in the user’s language, retrieve that content and respond in the user’s language, with citations explicitly mapping to the source languages used.
Cross-Language Synthesis
When content does not exist in the user’s language, retrieve English content, automatically translate and respond in the user’s language, and provide English citations only.
Guaranteed English Fallback
English shall always act as the unbreakable fallback content language ensuring all manuals are available to the LLM context limits globally.
Additional Platform Features
The integration natively extended their existing capabilities, bringing modern LLM interactions without massive custom engineering overhead.
- ✓ Fully Mobile/Desktop Responsive
- ✓ Dynamic Model Selector
- ✓ UI Data Source Selector
- ✓ Live Tokens In/Out Metrics
- ✓ Visible Reasoning Steps
- ✓ Floating Global Chat Widget
- ✓ Integrated Web Search Fallbacks
Immediate Business ROI & Benefits
Massive Consolidation
Consolidated huge swathes of data sources, distinct LLM models, discrete search indexes, and tooling into one unified Nextbrick application.
Removed Engineering Bottlenecks
Eliminated weeks of manual setup time for AI rules, Spark ingestion jobs, Kafka streaming jobs, and tedious QA processes.
Cost Eradication
Drastically cut down high recurring LLM API costs through local deployment strategies, eliminating external cloud reliance and associated software licensing overages.
Absolute Data Security
Guaranteed zero downtime and maintained 100% Secure Data standards by processing complex, proprietary queries totally on-premise.