EphraimAsad/BactAI-D
Hybrid AI System For Microbiology Phenotype Identification
What's novel
Hybrid AI System For Microbiology Phenotype Identification
Code Analysis
12 files read · 3 roundsAn AI-powered bacterial genus identification system that parses natural-language phenotypic descriptions using a tri-parser fusion (regex rules, extended biochemical, fine-tuned T5 LLM), scores candidates against a database and XGBoost model, applies diagnostic anchor overrides, and generates RAG-po
Strengths
The tri-parser fusion with learned per-field reliability weights, the 49KB rule parser handling dozens of microbiology phrasing variants, and the adaptive hybrid ranking (TF vs ML weighting based on confidence) represent genuinely deep domain-specific engineering. The architecture is clean with clear separation between parsing, scoring, ML prediction, RAG, and API layers.
Weaknesses
Test coverage is minimal (only a gold_tester and parser_eval in training/, no unit test suite visible), and the backend app.py is a monolithic 18KB file that mixes routing, business logic, and formatting. The LLM parser loads a full T5 model into memory with no graceful degradation if the model file is corrupt or too large for available RAM.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
6
Contributors
2
Files
562
Active weeks
1
Repository
Language
Python
Stars
1
Forks
0
License
—