[GSoC 2026] Introduction & Proposal for Continue AI-Powered Chatbot - Yugansh (@Yugansh5013)

Contributing GSoC
,
1

Hi Jenkins community,

I’m Yugansh (@Yugansh5013), applying for GSoC 2026: Continue AI-Powered Chatbot for Quick Access to Jenkins Resources (jenkinsci/resources-ai-chatbot-plugin).

Background

Contributing since December 2025 across the full stack — frontend, backend, performance, and tests:

5 merged PRs:

  • Export chat with multi-format download (#79)
  • Proactive log analysis & context observer for build failures (#89)
  • Fix BM25 OOM crash on module import (#108)
  • GSoC report: PDF → Markdown (#118)
  • Fix duplicate file attachments in Input component (#160)

5 PRs in review:

  • Jenkins authentication & secure user context (#106)
  • Replace LLM log extraction with deterministic regex parser (#148)
  • Connect missing log analysis logic to RAG pipeline (#232)
  • Expand prompt_builder unit tests — log_context branch & edge cases (#204)
  • Standardize input icons with lucide-react for dark mode (#267)

6 issues filed: dead log analysis code (#231), LLM extraction latency (#147), prompt_builder test gap (#203), duplicate file attachment bug (#159), dark mode input icons (#266), pytest mock warning (#139)

Full history: pulls?q=author:Yugansh5013

Proposal Direction: From RAG to Agentic Intelligence

My proposal focuses on evolving the chatbot into a production-ready Intelligent Diagnosis Agent through two core phases:

Phase 1: Graph RAG & Intelligent Build Failure Diagnosis

I aim to bridge the gap between static documentation and the live, interconnected Jenkins ecosystem:

  • Graph RAG Implementation: Moving beyond vector search by building a Knowledge Graph of Plugins, Versions, and Dependencies. This allows the LLM to traverse relationship edges to pinpoint complex dependency conflicts that standard RAG misses.
  • Intelligent Build Failure Diagnosis Agent: Building a multi-step reasoning agent that bypasses context limits via aggressive log chunking (extracting only ERROR/Stack traces) to synthesize root-cause analysis for failed builds.

Phase 2: Secure Context & Evaluative AI (LLM-as-a-Judge)

To make the plugin enterprise-ready and maintainable:

  • Jenkins Auth & Data Isolation: Finalizing the connection to the Jenkins Security Realm. I will implement data partitioning so users can only traverse Graph RAG nodes (logs/builds) they have explicit Job/Read permissions for.
  • Automated Evaluation (Ragas/TruLens): Integrating an “LLM-as-a-Judge” pipeline to score Answer Faithfulness and Context Precision, ensuring future prompt tweaks are backed by mathematical metrics rather than anecdotal checks.

Looking forward to feedback from mentors @krisstern and @berviantoleo.

One final question: What would be the preferred way to get feedback on the actual proposal, and where should I share it ?

2

Hey buddy! For getting the review for the proposal you can share it via this google form and make sure that, you make the link in such a way that anyone can view, who has the link which you’re submitting in the google form.
I think it helps.. feel free to ask queries if any…!

1 Like