We will be undergoing planned maintenance on January 16th, 2026 at 1:00pm UTC. Please make sure to save your work.

Opsmate

Opsmate automates project management for ops teams

Comment

Opsmate - DataOps JIRA Agent

Inspiration

We watched a healthcare DataOps PM spend 3 hours every morning creating JIRA tickets from email requests - many nearly identical to last week's, but requiring manual processing because patient data couldn't leave their servers.

The problem is universal for data teams:

  • 70% unplanned work scattered across emails, Slack, and meetings
  • 30% of their day lost to repetitive ticket creation
  • Zero automation options for teams handling PII/PHI data
  • No learning from patterns - same requests processed manually every time

They needed automation that was intelligent yet completely private.

What it does

Opsmate automatically transforms unstructured requests from emails, Slack messages, and documentation into ready-to-approve JIRA tickets in under 60 seconds.

Core capabilities:

  • Multi-source extraction: Monitors emails, Slack, Teams, and file drops
  • Local LLM processing: Uses Ollama-hosted models, keeping data on-premise
  • Intelligent scraping: Browser Use extracts knowledge from Confluence/wikis
  • PII/PHI protection: Automatic detection and masking for compliance
  • Pattern learning: Identifies recurring requests and applies templates
  • Human-in-the-loop: Full editing control before ticket creation

The system builds a growing knowledge base with every interaction, improving from 70% to 92% accuracy within weeks.

How we built it

We started by deploying Ollama locally to host Llama2 for text extraction without external API calls. Our Python backend monitors email inboxes (IMAP) and Slack channels (webhooks), feeding content through our extraction pipeline.

Technical implementation:

  • FastAPI backend for REST APIs and WebSocket connections
  • Browser Use integration for scraping JavaScript-heavy documentation sites
  • PostgreSQL + Qdrant for hybrid structured/vector storage
  • React frontend with Material-UI for the review interface
  • Docker containers for easy deployment and scaling
  • Redis for caching and real-time state management

Browser Use became our secret weapon - capturing documentation context from internal wikis that lacked APIs, dramatically improving extraction accuracy.

Challenges we ran into

Dynamic content extraction was our biggest hurdle. Internal documentation sites use heavy JavaScript rendering that traditional scrapers couldn't handle. Browser Use solved this by providing full browser automation capabilities.

Other key challenges:

  • PII detection balance: Too aggressive and we lost context, too lenient risked compliance
  • LLM hallucinations: Model invented plausible table names - fixed with schema validation
  • User trust: Initial "fully automated" approach was rejected - PMs needed control
  • Performance at scale: Local LLMs + hundreds of requests required careful optimization
  • Session management: Browser Use sessions for efficient repeated scraping

Accomplishments that we're proud of

We reduced ticket creation time from 10+ minutes to under 60 seconds while maintaining 100% data privacy - no information ever leaves the organization's servers.

Key achievements:

  • 92% extraction accuracy through Browser Use knowledge scraping
  • 100% PII/PHI compliance with automatic detection and masking
  • 15 hours saved weekly per PM on ticket creation
  • Pattern library that learns and improves with every correction
  • Zero external API calls - complete on-premise operation
  • Trust through transparency - full visibility and control for users

What we learned

Browser automation with Browser Use was transformative - 80% of organizational knowledge lives in wikis and documentation sites without APIs. This integration alone improved our accuracy by 20%.

Key insights:

  • Local LLMs are sufficient for extraction tasks - no need for GPT-4
  • Simple patterns beat complex AI for 80% of use cases
  • Human control builds trust - review and edit capabilities are essential
  • Context is everything - similar past tickets improved accuracy dramatically
  • Privacy-first architecture is non-negotiable for enterprise adoption

What's next for Opsmate

We're expanding Browser Use integration to scrape SharePoint, internal knowledge bases, and legacy systems that trap valuable context. The ability to interact with any web-based system opens unlimited possibilities.

Roadmap priorities:

  • Voice input for capturing requests directly from meetings
  • Shareable pattern library for collective intelligence across teams
  • Fine-tuned models specifically for ticket extraction
  • Predictive analytics to forecast request volumes
  • Multi-language support for global teams
  • Advanced Browser Use features: Form filling, multi-tab workflows, and complex navigation patterns

Our vision is to free DataOps teams from operational overhead, letting them focus on delivering value rather than managing tickets.

Built With

Share this project:

Updates