Role
Builder
Stack
PythonTensorFlow
Impact
Academic QA prototype
Problem
Academic QA experiments needed high-recall semantic retrieval.
Architecture
Python · TensorFlow · vector search.
Key decisions
- Pre-LLM era - semantic embeddings were the right primitive.
Outcome
High-recall knowledge lookups for academic QA.
What I learned
Embeddings are still the right hammer for many nails.
What's next
Hybrid retrieval with LLM reranking.
Next
Browse more selected work