SaaS
Cortex Support Agent
Resolved 71% of support tickets autonomously with an LLM agent.
- Timeline
- 10 weeks
- Team
- 3 engineers
- Status
- Live
- Client
- SaaS
- Completed
- 2025
- Services
- AI Automation, AI Integration, API Development
Executive summary
The outcome, first.
Cortex's support team was overwhelmed by repetitive tickets, slowing response times and burning out agents.
NWARRAH built an LLM support agent that resolves common issues end-to-end and hands off cleanly to humans when needed.
Support volume dropped while satisfaction rose, and human agents now focus on complex, high-value conversations.
Client challenge
The problem
Repetitive tickets were drowning a talented support team.
- High volume of repetitive, low-complexity tickets
- Slow first-response times during peak hours
- Knowledge scattered across docs and past tickets
- No safe way to automate without risking bad answers
- Agent burnout from monotonous work
Discovery
How we de-risked it
We analyzed historical tickets to find what could be safely automated.
- Clustered tickets by intent and resolution path
- Identified safe, high-volume automation candidates
- Designed retrieval over docs and resolved tickets
- Defined guardrails and escalation criteria
- Planned evaluation to measure answer quality
Solution
How NWARRAH solved it.
A retrieval-grounded agent with strong guardrails and clean human handoff.
Grounded answers
Responses cite your docs and past resolutions.
Safe guardrails
Escalates when uncertain instead of guessing.
Continuous evals
Answer quality is measured and improved.
Seamless handoff
Context passes to humans with full history.
System architecture
Engineered to last.
A clean, observable data flow from interface to infrastructure.
Agent Console
Support UI with AI suggestions and handoff.
API Layer
Orchestrates retrieval and model calls.
LLM Agent
Retrieval-grounded reasoning with guardrails.
Vector + PostgreSQL
Knowledge index and conversation store.
Automation
Auto-resolution and escalation routing.
Analytics & Evals
Quality metrics and knowledge tuning.
Features
What we shipped.
AI agent
Autonomous resolution of common tickets.
Search
Semantic search over knowledge base.
Real-time
Instant suggestions for human agents.
Analytics
Resolution and satisfaction metrics.
Notifications
Escalation alerts to the right team.
Admin panel
Manage knowledge and guardrails.
Concept artifacts
How the system fits together.
These are illustrative engineering artifacts of the delivered system. Live product screenshots are published here as each client approves them.
Business results
Measurable impact.
Tickets auto-resolved
Faster first response
Lower support cost
Higher CSAT
"It feels like we hired a tireless senior agent. The handoff is so clean our team barely notices where the AI ends and they begin."
Daniel Osei
Head of Support, Cortex
Engineering insights
Decisions that mattered.
Retrieval grounding
Every answer is grounded in source content to minimize hallucination.
Evaluation harness
A golden dataset gates prompt and model changes before they ship.
Confidence routing
Low-confidence cases escalate automatically with full context.
Let's build something together.
Tell us the outcome you need. We'll engineer the system that gets you there.
