Tirth Kanani is an AI and automation engineer at Telana who researches how to give language models the context a task needs. He created code-review-graph, a local structural index for coding assistants.[1][2]
Start with code-review-graph →
X · tirth_8205 Career
Kanani’s MSc research at the University of Birmingham explored local language models that answer questions about private documents and trace answers to their sources. His later work includes auditable agents for interactive reasoning and tools that turn a codebase or a product interaction into information an assistant can use. He describes his engineering role at Telana as AI and automation work.[1]
Contributions
A map of code relationships
Code-review-graph parses a repository with Tree-sitter and maintains a structural graph incrementally as files change. Through MCP, a coding assistant can request context around a change instead of repeatedly reading large parts of the repository. The graph connects code elements and their relationships, giving review a starting map of the affected area. This narrows retrieval; the assistant still has to examine and reason about the relevant code.[2]
Reviewing a product in use
Kanani is also building CrumbleUX, which examines product interactions from six perspectives. His broader research asks how an agent’s decisions can be inspected afterward, including work evaluated against interactive abstract-reasoning tasks.[1]
Tools, writing & talks
Tools
Start here
code-review-graph
github.comMaintains a local structural code graph and serves relevant review context to coding agents over MCP.
Read the before-and-after context example, then build the graph for a small repository you understand. Running the tool requires Python and a compatible coding client for its MCP integration.
Creator and maintainer[1][2]Original source