AI Engineer · Retrieval · Agents · LLM Post-Training
I build AI systems around retrieval, recommendation, and language models.
Currently studying Computer Science at USC. Before diving deeper into agents and LLMs, I spent several years building large-scale recommendation and retrieval systems.
|
Agentic Retrieval for Knowledge Bases An agentic retrieval system where the model decides how to search, what to read, and when to stop, instead of following a fixed RAG pipeline.
BM25 · semantic · hybrid · RRF · reranking |
Small experiments, one idea at a time A learning lab for understanding LLM post-training through small, reproducible experiments on Apple Silicon. Currently: SFT + LoRA with MLX |
Retrieval + Agents
- How should an agent decide which retrieval tool to use?
- When should it search again, read deeper, or stop?
- How should agentic retrieval be evaluated beyond final-answer accuracy?
- Can retrieval strategies themselves be improved through post-training?
LLM Post-Training
- SFT and instruction tuning
- Preference optimization
- Online RL and GRPO
- Reward design for tool-using agents
- Efficient post-training of small open-weight models
Recommendation + Search
- Retrieval and ranking
- Two-tower models
- Multi-interest representation learning
- Real-time recommendation systems
- The overlap between classic RecSys retrieval and modern agentic search
LLM / ML: Transformers · MLX · Ollama
Retrieval: BM25 · Qdrant · RRF · neural reranking
Data / Systems: Spark · Flink · Kafka
I keep longer notes, project write-ups, and experiments at bogao.dev.
build → measure → understand → repeat