An AI/ML engineer (still learning some things) and full‑stack developer fueled primarily by caffeine ☕.
- Fixed a critical bug in Hugging Face Transformers: filtered out
Nonerouter logits when usingmlp_only_layerswithoutput_router_logits=True. The fix was accepted and merged as PR [#39206], resolving issue [#39203]. - Deepened expertise with MoE model architecture, auxiliary loss functions, and advanced inference logic.
Built a robust platform to streamline academic content creation:
- Tech Stack: React, TypeScript, Node.js, Supabase, Clerk, Google Gemini, Tailwind & Shadcn UI
- AI Logic: Generated structured outlines from documents + rubrics; extracted academic constraints (citation style, word count) via AI
- Backend: Supabase authentication + storage (signed URLs + RLS), Express APIs, secure file handling
- Frontend: Multi-step wizard with interactive forms, PDF handling, resilient edge-case management
- Contributed to production-grade CI/CD pipelines with Docker, Kubernetes, and ArgoCD
- Managed diverse YAML configs for deployment, secrets, autoscaling, and resource allocation
- Designed and maintained infrastructure on AWS (EKS, ECR, IAM, S3, RDS) for scalable service delivery
- Languages: TypeScript, Python, Kotlin (learning)
- Frontend: React, Vite, Tailwind, Shadcn UI
- Backend: Node.js, Express.js
- Database: Supabase (PostgreSQL, Storage, Auth, RLS)
- Auth & Storage: Clerk authentication, Supabase storage (signed URLs)
- AI Systems: Google Gemini, Hugging Face Transformers
- ML Dev: MoE modeling, Transformer pipelines, caching, performance tuning
- DevOps & CI/CD: Docker, Kubernetes, ArgoCD, AWS EKS/ECR/RDS/S3, GitHub Actions, Ruff, pytest
- Other Tools: pdfjs‑dist, Multer, REST APIs, YAML config management
- Discord:
swlft_ - Pronouns: He/Him
- Fun Fact: I have a love-hate relationship with Python—but still use it daily! 😄
Thanks for stopping by! Feel free to check out my pinned repositories and open-source contributions.