Traditional org charts fail to capture how work actually gets done, but graph models can reveal the hidden network of collaboration.
The initial graph is built from the formal org chart, mapping out the direct reporting structure of a sample company.
Augment the org chart with communication data from tools like Slack or surveys to identify bottlenecks and cross-departmental interactions.
Adding service ownership data from tools like Backstage helps visualize team workload balance, dependencies, and informs architectural decisions via Conway's Law.
Parsing pull request data from source control reveals actual code ownership, technical bridges between teams, and domain experts.
Further enrich the organizational graph by incorporating data from trust surveys, documentation tools like Notion, and work item trackers like Jira.
Apply the Leiden community detection algorithm to find densely connected clusters of people or services, revealing hidden teams or monolith decomposition paths.
The betweenness centrality algorithm identifies nodes that connect different parts of the graph, highlighting key individuals who act as either crucial bridges or bottlenecks.
Use the PageRank algorithm to rank the influence of people or services based on their incoming connections, pinpointing critical points in the system.
Connect an LLM to the graph database to ask high-level, natural language questions about risks, bottlenecks, and optimal team composition for projects.
Follow a step-by-step process for implementing this analysis, from forming a hypothesis and collecting data to importing it into Neo4j and experimenting.
This data should be used to understand system dynamics, not for performance evaluation, requiring strict confidentiality and careful expectation management.
Jobs that call for the skills explored in this talk.