Inspiration
Every developer knows how difficult it can be to work with an unfamiliar or unfinished codebase. There may be unresolved issues, incomplete features, confusing dependencies and no clear path to completion.
I created QEDA to address this problem. QEDA means finish, and that idea is central to the platform: helping developers move from uncertainty to informed, evidence-backed software recovery.
What it does
QEDA is an AI-powered engineering workspace that helps developers understand, investigate, recover and complete existing software projects.
Developers begin by uploading a repository as a ZIP file. QEDA analyzes its structure, identifies technologies, examines source files and highlights potential issues.
Its integrated assistant, Palm AI, uses NVIDIA Nemotron through Nebius Token Factory to provide contextual explanations and engineering recommendations based on actual repository content.
QEDA brings together several connected capabilities:
- AI Patch Studio: Generates targeted code improvements with original-versus-proposed comparisons and developer approval.
- Verification Intelligence: Checks supported Python and JSON syntax and records genuine validation results.
- Project Autopsy: Produces evidence-based completion plans.
- Recovery Centre: Organizes findings into actionable recovery tasks.
- RETYRE: Investigates declared dependencies and possible source-code usage.
- REKOVA: Connects recovery cases with engineering tasks, patches and verification evidence.
- Debug Journey: Preserves a traceable history of project investigations and development decisions.
- Proof of Finish: Generates evidence reports that identify progress, outstanding work and verification limitations.
QEDA is designed to support developer decisions without falsely claiming that a proposed patch has been applied or that syntax validation proves runtime correctness.
How I built it
I developed QEDA using React, TypeScript and Vite for the frontend, with Python and FastAPI powering the backend.
I integrated NVIDIA Nemotron through the Nebius Token Factory inference API to power Palm AI and targeted patch generation.
The backend performs repository analysis, retrieves relevant source-code context and manages persistent local development records through SQLite.
Supabase provides authentication, while account-level authorization helps ensure that users can only access their own repositories and engineering records.
The application is deployed using Vercel for the frontend and Render for the backend.
Challenges I ran into
One of the biggest challenges was generating meaningful code changes without unnecessarily modifying unrelated functionality.
I addressed this by developing a source-selection workflow that provides relevant code sections to Nemotron, reconstructs the complete proposed file and validates supported syntax.
Other challenges included securing repository uploads, implementing cross-account isolation and maintaining traceable evidence across multiple engineering workflows.
Accomplishments I'm proud of
I'm proud of building a functional platform that goes beyond a standalone AI chatbot.
QEDA connects real repository analysis, contextual AI assistance, targeted patch generation, dependency intelligence and evidence-based recovery planning.
I also successfully deployed the application and tested repository uploads, Palm AI, AI Patch Studio, Verification Intelligence, RETYRE, Project Autopsy, Recovery Centre, REKOVA and Proof of Finish on the live website.
What I learned
Building QEDA strengthened my understanding of contextual AI systems, NVIDIA model integration, full-stack development, API security and engineering workflows that make AI-generated results more transparent.
It also taught me the importance of distinguishing observed evidence from assumptions and keeping developers in control of important decisions.
What's next for QEDA
My next goal is to evolve QEDA from an AI-powered code recovery workspace into a more autonomous, end-to-end software repair platform.
I plan to introduce AI-Powered Repository Repair and ZIP Export, allowing developers to upload an existing project, let QEDA investigate its issues, generate and apply approved fixes to a separate working copy, and download a new ZIP containing the updated project.
This will include automated repair planning, multi-file code improvements, isolated runtime testing, AI-assisted debugging, and controlled repair iterations when tests fail.
Developers will be able to review exactly which files were changed, compare the original and repaired versions, inspect test results, and approve the final changes before downloading their updated project.
I also plan to expand QEDA with GitHub repository integration, additional programming-language support, persistent cloud storage, and more advanced dependency intelligence.
The long-term vision is to make QEDA a complete engineering recovery platform that doesn't just explain what's wrong with a project, but helps developers repair it, verify the results, and take home a working, improved version of their original codebase.
Log in or sign up for Devpost to join the conversation.