Hi all, My name is Koushik Bhargav Nimoji, with the UNH-IOL. I would like to introduce an opportunity for an agentic development tool catered to the DPDK project with a focus on minimizing token costs. This is a project that I have created myself In my own free time. I'm sharing this technical brief to see if there is any interest in the community for further development.
======================================================================== Overview: DTS-Helper is an agentic development tool, built to accelerate developer workflows through a custom RAG pipeline and automated dual-agent validation loop. By eliminating brute-force LLM codebase crawls and context bloating, token costs are reduced. Advantages: - Cost-efficient: Removes the need for an agent to crawl the codebase which fills the context window and is a token-heavy approach. - Deterministic: Provides deterministic code indexing and retrieval, opposed to manual agentic code searching. - Code quality assurance: Unlike AI tools that dump untested code into a workspace, DTS-Helper uses a secondary validation agent to assess changes and provide a new perspective to catch logical flaws and syntax errors. - Seamless Integration: DTS-Helper fits into the DPDK workflow as a development assistant, as the output format of any changes is a patch file located in the output directory Capabilities: - Pulls the DPDK repo, then chunks and stores DTS code into a local ChromaDB vector store - The user can then ask the tool a specific question about the codebase, to which the tool can query the vector store to gain context behind the question before answering - The user can also request a code change, to which a temporary workspace is created. The agent then makes informed code changes in that temporary workspace. - Once the development agent is finished, a validation agent reviews the changes. If the changes are invalid, the validation agent sends feedback to the development agent on what specifically to fix. If the changes are valid, a patch file is then created and placed in the output directory. Next Steps: - Expanding vector store to hold the DPDK codebase for full-repo coverage or DTS test plans for testsuite migrations. - Provide the tool with a concrete skills file that contains coding conventions, etc. - Simplify user configuration for AI providers/models - In development: Enable the validation agent to run the dts-check-format.sh script Repository: https://github.com/Koush1/agentic-dts The readme in the repository contains a workflow diagram of the tool. ======================================================================== Please reach out with any comments or thoughts on this, and any feedback would be greatly appreciated. Best, Koushik Bhargav Nimoji
