VoiceOps Agent

Context-driven voice agent that converts commands into verified actions - Jira tickets, Slack notifications, and full audit trails powered by Elasticsearch Agent Builder.

Comment

Inspiration

My daily workflow was exhausting. Constantly switching between Jira to create tickets, GitHub to track issues, and Slack to notify the team - it felt like I was spending more time on tooling than actual work.

I thought: What if I could just speak what I need and let an AI agent handle the rest?

That's when I decided to build VoiceOps Agent - a context-driven AI agent powered by Elasticsearch that transforms voice commands into real actions. Now I speak, and everything works in seconds.

What it does

VoiceOps Agent lets you speak naturally to automate operations tasks:

"Create a high-priority ticket for the login page freezing on Safari, assign to backend, and notify Slack"

The agent then:

  1. Extracts intent - Identifies the action, priority, team, and details
  2. Retrieves context - Searches Elasticsearch for similar tickets and past commands
  3. Reasons about the best action - Chooses the right project, assignee, and tools based on historical data
  4. Requests confirmation - Shows its reasoning and asks for approval before executing
  5. Executes and logs - Creates the Jira ticket, sends Slack notification, and logs everything to Elasticsearch

The result? What used to take 3+ minutes now takes 8 seconds.

How I built it

Backend (FastAPI + Python):

  • Elasticsearch for storing tickets, commands, and audit logs
  • ES|QL queries for real-time analytics
  • Groq LLaMA 3.3 70B for intent extraction and reasoning
  • Jira REST API and Slack Webhooks for real integrations

Frontend (React + TypeScript):

  • Voice recording with Web Audio API
  • Real-time pipeline visualization showing agent reasoning
  • ES|QL Explorer for live query execution
  • Impact dashboard showing time saved

Elasticsearch Agent Builder:

  • Multi-step workflows with tool orchestration
  • Context retrieval using semantic search
  • Full audit logging for compliance

Challenges I ran into

1. Balancing autonomy with safety I didn't want the agent to blindly execute actions. I implemented a confirmation flow where users review the agent's reasoning before approving execution.

2. Handling voice transcription errors Speech-to-text isn't perfect. I built fuzzy matching and context-aware intent extraction to handle variations and errors gracefully.

3. ES|QL syntax learning curve Some ES|QL functions work differently than expected. For example, DATE_FORMAT vs DATE_TRUNC for time-series grouping required debugging.

Accomplishments that I'm proud of

  • 22x faster ticket creation (3 minutes → 8 seconds)
  • Real integrations - actual Jira tickets and Slack notifications
  • ES|QL Explorer - live queries against Elasticsearch
  • Full audit trail - every action logged with reasoning
  • Context-driven decisions - agent uses historical data, not just prompts

What I learned

  • ES|QL is incredibly powerful for real-time analytics
  • Multi-step agents need transparency - showing reasoning builds trust
  • Elasticsearch's search relevance works surprisingly well out-of-the-box
  • Voice interfaces need confirmation flows for safety

What's next for VoiceOps Agent

  • GitLab integration - create issues and link to merge requests
  • Multi-agent architecture - separate planning, execution, and review agents
  • Mobile app - voice commands on the go

Built With

Share this project:

Updates