Confluence Copilot Cover Page
Projec Task Assignment Visualization
Project Overview and List of Employee/Tasks Page
Employee Details
Magic Ask for Employee Interaction Graph
Magic Ask for Task Dependence Graph
Default View for Task Dependence Graph Visualization
Magic View for Intuitive Task Dependence Graph Visualization
Employee Promotion Profiling with HITS
Task Prioritization System with CPM

Confluence Copilot

Efficient project and HR management GraphRAG Agent with ArangoDB

Problem

Employee Frustration & Time Wastage
Inefficient task management within corporate workflows results in wasted time and value. This affects employee productivity and morale.
Lack of Integration
Current tools like JIRA, Azure DevOps, and Atlassian Confluence are fragmented, leading to poor integration and inefficiencies in managing tasks and resources.
Underutilization of Project Management Frameworks
Companies fail to fully leverage project management frameworks, leaving potential efficiencies on the table.
Lack of Observability
There's limited visibility into task progress and human resource allocation, which leads to poor decision-making and missed opportunities.
Employee Contribution & Promotion Issues
Lack of attention to individual contributions results in missed promotions and high employee turnover. Promotions are often subjective and fail to account for every employee's impact.

Some Statistics

Time Wastage:
Employees lose 26% of their day to avoidable administrative tasks, outdated processes, and tool fragmentation, equating to 76 working days per year wasted on inefficiencies.
Burnout & Stress:
83% of professionals report feeling overwhelmed due to unclear workflows and repetitive manual tasks.
Employee Turnover:
Employees promoted without proper preparation are 29% more likely to leave within a month compared to those who were not promoted.

Solution

Introducing Confluence Copilot

We introduce an AI-powered agent platform that gains contextual understanding of corporate projects and human resources by integrating popular workflow tools like JIRA and Azure DevOps into a unified knowledge base.

How It Works:

We convert these tools into graph representations and store them in ArangoDB.
By using GraphRAG and Graph Databases, we provide real-time context and suggestions to HR and Team Leads, improving decision-making on project timelines and employee promotions.

Dataset

To demonstrate our agent, we synthesized real-world data based on how tasks and employees are distributed within a company.

Company D

Industry: Data Engineering
Employees: 36 employees, divided into 4 teams:
- Business Intelligence
- Data Science
- Data Governance
- Data Engineering
Projects:
- StreamSync Pipeline (90 tasks)
- DataForge ETL (44 tasks)
- AetherFlow Orchestrator (49 tasks)
- NeoGraph Linker (45 tasks)

These teams use Azure DevOps and JIRA to track tasks and progress.

Graph Representation:

To create real-time insights and optimize HR processes, we converted the data into three types of graphs:

Employee Interaction Graph:
Represents the interactions among employees (e.g., task reviews, advising, collaboration).
Task Dependency Graph:
Represents the relationships between tasks, showing how subsequent tasks depend on the completion of prior tasks.
Task Assignment Graph:
Represents the allocation of employees to tasks, factoring in availability through the Critical Path Method (CPM) and HITS (Hyperlink-Induced Topic Search) algorithms.

These graphs are built on top of existing task databases (JIRA and Azure DevOps) and the company’s personnel database.

Technical Architecture

Frontend: Streamlit for user interfaces that display the dynamic graphs and agent interactions.
Backend:
- ReAct Agent: An agent powered by GraphRAG to process real-time project and resource data.
- ArangoDB: A multi-model database that stores task, dependency, and employee interaction data in graph format.
Key Algorithms:
- Critical Path Method (CPM): Helps allocate human resources and calculate slack time for efficient project management.
- HITS Algorithm: Used to evaluate employee contributions and potential for promotion based on authority and hub scores.

Challenges We Encountered

Data Availability:
Since we were exploring a new problem space, existing data was scarce. We overcame this by synthesizing data from real-world task distributions and employee workflows.
Graph Representation Complexity:
Graph representation of tasks and employee interactions requires complex computations. We used NVIDIA cuGraph integrated with ArangoDB to accelerate time-intensive graph algorithms like CPM and HITS, making the solution scalable for real-time data.
ReAct Agent Complexity:
Enhancing the ReAct Agent with dynamic AgentState to create graphs and data structures was a challenging endeavor. Most agents work with string-based messages, but we expanded this to handle real-time graph computation, which required deep expertise in analytics workflows and OOP.

Accomplishments We're Proud Of

Dynamic Graph View:
We successfully built a scalable and production-ready dynamic graph view that is intuitive and usable in real-time for both HR and Team Leads.
Efficient Computational Performance:
Integrating NVIDIA cuGraph into ArangoDB significantly boosted the performance of time-intensive algorithms (CPM, HITS), making them suitable for processing large datasets on the fly.
ReAct Agent Success:
Our ReAct Agent demonstrated incredible accuracy, navigating complex global environments to suggest resource allocation and evaluate employee contributions.

What We Learned

State-of-the-Art GraphRAG Technologies:
We deepened our understanding of Graph-based AI technologies, particularly in managing corporate tasks and HR processes.
Research & Development:
The project significantly improved our research abilities, especially in graph analytics and AI agents. A big shout-out to my teammate, a true 10x engineer!
ReAct Agent Framework:
We gained valuable experience in building and scaling ReAct Agents, and how they can be extended to handle complex real-time data.
Efficient Development Practices:
We learned the importance of committing small, committing fast to maintain momentum and progress.

What's Next for Confluence Copilot

Productization:
We plan to make Confluence Copilot an official product by enhancing its scalability and adding more features.
Probabilistic Models:
We aim to integrate probabilistic models to improve the forecasting capabilities of CPM and make more accurate predictions.
Continued Optimization:
Further optimization of the NVIDIA cuGraph and ArangoDB integration to handle even larger datasets and improve performance.