薛瑞锋
I build AI applications, backend systems, and developer tools—and contribute to the open-source projects that developers use every day.
Building AI applications and open-source projects
> whoami
Ruifeng Xue
> focus
AI Applications
Backend Engineering
Open Source
>
01 / SELECTED PROJECTS
LLM-powered Financial Research
A stock analysis and research reporting system combining market data, sentiment, machine learning, and LLMs. Deployed on Tencent Cloud Linux.
AI storytelling & video workspace
Turn ideas into branching stories and generated videos with Qwen, Wan, and version management that preserves each story path.
AI Coding Configuration & Collaboration
Contributing to Tencent’s team-level AI Coding configuration tool: managing Skills, Rules, and MCP configurations across multiple coding tools.
Find open-source issues worth contributing to
Search and score open GitHub issues by language, labels, and activity to help developers discover approachable contributions.
02 / INTERNSHIP
AI applications and Python backends, from model integration to reliable generation services and asset delivery.
Seele AI, Inc.
Models cleaned per day
Faster than manual review
Average game generation time
Assets in the business library
Contributed to uvGen / imageGen, integrating GPT, Gemini, Hunyuan, Trellis2, and Hitem3D. Adapted request and response formats for text-to-image, image editing, and text/image-to-3D generation.
Developed business APIs for assets, generation tasks, review, and repair. Connected model calls, Workflow, image processing, and S3 uploads; helped integrate the service gateway and prepare test environments.
Improved input validation, exception handling, and retries, and extended Consul configuration. Investigated uvGen memory issues and Gemini timeouts with Grafana / Loki logs and Feishu alerts.
Used Blender bpy for compression, size normalization, FBX-to-GLB conversion, and post-processing. Fixed metallic/roughness texture loss and alpha-channel issues for Unity and Three.js.
Applied VLMs to part semantics, material classification, and labeling. Extended skeleton recognition and asset review/repair APIs, and documented quality rules and error codes.
Helped clean a 10M-scale asset library, filter untextured assets, and automate image compression and format processing. Worked with algorithm, art, and frontend teams to validate generation quality and improve the delivery workflow.
From integration to delivery
Investigated texture loss during import and repaired material mapping for downstream engines.
Investigated abnormal memory usage in the generation service using production logs and task data.
Traced timeout responses and improved exception handling and retry behavior in model calls.
Repaired alpha-channel handling and material node connections in the Blender processing pipeline.
Extended repair endpoints for Unity, Blender, and Three.js asset compatibility issues.
Helped prepare asset-platform test environments and document pipeline error codes for integration and troubleshooting.
03 / OPEN SOURCE
Over the last 6 months
Contributing code and joining technical discussions on GitHub, collaborating on—and helping maintain—medium and large open-source projects.
Contributions include projects from Tencent, Google, Anthropic, OpenAI, and their open-source communities.
Explore my GitHubOctober 2026
Added Router::add_service and route composition APIs for top-down multi-service registration, preserving the existing registration interface.
View PR ↗Made pull remove deleted team documents, continue syncing other resources after failures, and clear version markers so the next pull can retry.
View PR ↗Normalized command names across platforms and blocked interpreters and command executors from bypassing the security policy through custom allowlists.
View PR ↗Added descriptor_set_bytes to accept in-memory descriptor data without temporary files, preserving the file-based API.
View PR ↗Refreshed unchanged stale Copilot rules and applyTo configuration before push, while preserving real local edits.
View PR ↗Added wheel scaling for the targeted ModelViewer, normalized wheel delta units, and preserved page scrolling when no model is targeted.
View PR ↗Fixed the cursor staying at the insertion start after pasting from VS Code. Adjusted CodeMirror selection handling and added regression tests.
View PR ↗Added one-click JSON copying to the CVE conversion page, keeping the button visible while scrolling, with loading-state control and copy-success feedback.
View PR ↗Centralized 83 duplicated test-path setups in pytest conftest.py, unified imports, and reduced boilerplate while preserving test compatibility.
View PR ↗04 / TECH STACK
05 / EDUCATION
BSc in Computer Science and Technology · Full-time
06 / ABOUT
A long-term interest in AI, software engineering, and products that solve useful problems.
I learn by building real projects, working through engineering challenges, and collaborating in open source.