Liyihan11unique@outlook.com | LinkedIn | Personal Website
Languages
- Kotlin / Java / Python / C / SQL
AI & ML
- sciket-learn / ONNX / sherpa-ncnn / ASR (Zipformer)
- Model Optimization / Quantization
Development
- Android / Gradle / Jetpack Compose / Kotlin Multiplatform
- Backend basics / API integration
https://github.com/YihanLi-erisaer/smeeting
- On-device speech recognition and local AI summary system
- Kotlin + sherpa-ncnn + ncnn_llm deployment
performance on a android device Helio G99 (CPU) processor using armv8 libs only running ASR model
| Metric | Value |
|---|---|
| Memory Usage | ~250 MB |
| Latency | ~120 ms |
| Chinese Accuracy (in chaos environment) | ~89% |
| Chinese Accuracy (in quiet environment) | ~95% |
| English Accuracy (in chaos environment) | ~92% |
| English Accuracy (in quiet environment) | ~97% |
performance on a android device Helio G99 (CPU) processor running Qwen3.0_0.6b model
| Metric | Value |
|---|---|
| Memory Usage | ~3 GB |
| Throughput | ~2 token/s |
- Future change back to raw ncnn inference framework, and deploy a small LLM that can conclude the recognition text.
https://github.com/YihanLi-erisaer/aiclient
- AI API integration client
- Modular design for LLM usage
- KMP multiplatform project (Android + web)
- Focus on usability + extensibility
- local deloyment of asr model using ONNX Runtime framework
https://github.com/YihanLi-erisaer/Elements-of-data-processing-webpage-crawler
- Data collection pipeline
- Parsing + structured storage
- Useful for LLM / RAG preprocessing