Context-Aware Blocker Detection for Engineering Teams

The Problem

A Series A engineering team was scaling fast—too fast. Their Slack channels were a maze of threaded discussions, their Jira boards a graveyard of stale tickets. Engineering managers spent Sunday evenings grooming backlogs instead of preparing for the week. Blockers surfaced at standups when it was already too late.

The Architecture

flowchart LR
  subgraph sources [Data Sources]
      Slack[Slack Channels]
      Jira[Jira Board]
      Git[Git Activity]
  end
  
  subgraph agents [Agent Crew]
      Sentinel[Sentinel Agent]
      Sentiment[Sentiment Analyzer]
      Dependency[Dependency Mapper]
      Reporter[Report Generator]
  end
  
  subgraph outputs [Outputs]
      Alerts[Proactive Alerts]
      Dashboard[Manager Dashboard]
      StandupDoc[Standup Brief]
  end
  
  Slack --> Sentinel
  Slack --> Sentiment
  Jira --> Sentinel
  Jira --> Dependency
  Git --> Dependency
  
  Sentinel --> Alerts
  Sentiment -->|"Frustration Signals"| Sentinel
  Dependency -->|"Bottlenecks"| Sentinel
  Sentinel --> Reporter
  Reporter --> Dashboard
  Reporter --> StandupDoc

The Context-Aware Blocker Agent

The system deploys a crew of specialized agents:

1. Sentinel Agent: Monitors Jira status transitions and identifies tickets stuck in “In Progress” beyond normal cycle time
2. Sentiment Analyzer: Scans Slack threads for frustration signals—keywords, emoji patterns, and conversation tone shifts
3. Dependency Mapper: Traces ticket relationships and git branch activity to surface hidden technical debt
4. Report Generator: Synthesizes findings into actionable standup briefs and weekly reports

The magic happens when these agents collaborate—a Slack thread expressing frustration + a stale Jira ticket + blocked git branches = proactive alert before standup.

Tech Stack

• CrewAI — Multi-agent task orchestration
• GPT-4o — Natural language understanding and report generation
• Jira API — Real-time board state and ticket history
• Slack SDK — Channel monitoring and sentiment extraction
• Node.js — Event-driven backend

The Impact

Managers get a standup brief from Slack, Jira, and git activity instead of reconstructing the week by hand. Stuck tickets and frustrated threads surface before the meeting.