TopVibeCoder
Independent
A platform that helps everyone become a better vibe coder using LLM-as-judge over 3,000+ generated apps; published the TopVibeCoder Bench paper benchmarking the era of vibe coding.
Portfolio
Junfan Huang
2019
Train AI
2022
Speed up AI
2026
Vibe coders, assemble!
Now building
TopVibeCoder
Active
Previously at
-
-
Content2Learn
Independent
A platform that helps you learn and improve memory retention using a personalised style.
-
Hardware-Aware LLM Benchmarking Platform
KRAI Inc
Built an end-to-end benchmarking platform from scratch
-
Code Copilot Fine-tuning (LoRA)
KRAI Inc
Trained Code Copilot for internal tool
-
SDXL Inference
KRAI Inc
Made Stable Diffusion XL inference 3x faster on Qualcomm A100
-
Gaelic Spelling Tool (NLP)
University of Edinburgh
Launched a transformer-based orthographic standardiser for Scottish Gaelic.
-
ISGverse App
University of Cambridge
Built R/Shiny and Neo4j graph web apps to visualise IFN-regulated genes and biological pathways.
-
COBRA: TF targets hunter
University of Cambridge
Built a state-of-the-art ML Python package integrating multi-omics data to identify transcriptional targets at 92% precision
-
ChinHubs (second-hand marketplace)
ChinHubs
Co-founder & CTO. Led the team to build and launch a second-hand e-commerce platform in one month; scaled the team from 3 to 15, secured GBP 11,000 from Cambridge Downing Enterprise, and reached 3K+ registered users.
-
Instructor
University of Cambridge
Co-instructed Bioinformatics Summer School - taught 45 students
-
ML for Source-Code Vulnerability Detection
QinetiQ
Explored AST and code-property-graph representations and implemented ILP and CNN models to detect buffer overflows across 21,502 C/C++ scripts
-
OWASP Seraphimdroid (Malware Detection)
GSoC contributor 2019
GSoC student with the OWASP Foundation: deployed deep learning for real-time Android malware detection, using LSTM/Autoencoder for anomaly detection and CNN/Random Forests for classification (published paper).
