How joannapreethi28/icc-26 works
Make AI Know Her is a multilingual cricket-answer layer that intervenes on gender-neutral questions: it classifies the query, applies a policy, retrieves verified or computed facts, then renders a sourced answer. Users reach the shared resolver through REST/MCP or the demo UI. The repository also includes model training and classifier evaluation. Some links are documented only at the pipeline level; unsampled implementation details are not inferred.
Main entry points: pipeline.py Resolution pipeline · understand.py Query parser · facts.py Fact retrieval · service.py Validated service
Components 20
Serving
- REST and MCP
src/mak/api/app.py - Demo UI
src/mak/ui/demo.py - Validated service
src/mak/api/service.py - Resolution pipeline
src/mak/pipeline.py
Query understanding
- Query parser
src/mak/nlu/understand.py - Rules classifier
src/mak/nlu/rules.py - Laya classifier
src/mak/nlu/laya_head.py - Entity matching
src/mak/nlu/entities.py - People registry
src/mak/registry/people.py - Wikidata registry
src/mak/registry/wikidata.py
Facts and answers
- Decision policy
src/mak/policy/decide.py - Intent catalogue
src/mak/fetch/catalogue.py - Fact retrieval
src/mak/fetch/facts.py - Verified records
src/mak/records/golden.py - Computed records
src/mak/records/compute.py - Leader lookup
src/mak/records/leader.py - Answer rendering
src/mak/compose/render.py
Model and evaluation
- Laya fine-tuning
training/finetune/train_laya.py - Classifier evaluation
src/mak/eval/classifier_eval.py
Other components
- Questioner
Connections 23
- Questioner → REST and MCP: submits query
- Questioner → Demo UI: uses demo
- REST and MCP → Validated service: routes requests
src/mak/api/app.py - Demo UI → Validated service: calls resolver
src/mak/ui/demo.py - Validated service → Resolution pipeline: resolves query
src/mak/api/service.py - Resolution pipeline → Query parser: parses query
src/mak/pipeline.py - Resolution pipeline → Intent catalogue: looks up intent
src/mak/pipeline.py - Resolution pipeline → Decision policy: decides intervention
src/mak/pipeline.py - Resolution pipeline → Fact retrieval: fetches facts
src/mak/pipeline.py - Resolution pipeline → Answer rendering: renders answer
src/mak/pipeline.py - Query parser → Rules classifier: uses rules
src/mak/nlu/understand.py - Query parser → Laya classifier: uses model
src/mak/nlu/understand.py - Query parser → Entity matching: matches entities
src/mak/pipeline.py - Entity matching → People registry: looks up names
src/mak/nlu/entities.py - Fact retrieval → Verified records: loads verified facts
src/mak/fetch/facts.py - Fact retrieval → Computed records: uses computed records
src/mak/fetch/facts.py - Fact retrieval → People registry: uses player registry
src/mak/fetch/facts.py - Verified records → Leader lookup: loads leaders
src/mak/records/golden.py - Answer rendering → Leader lookup: uses leader data
src/mak/compose/render.py - Wikidata registry → People registry: updates registry
src/mak/registry/wikidata.py - Laya fine-tuning → Laya classifier: trains classifier
training/finetune/train_laya.py - Classifier evaluation → Rules classifier: evaluates classifier
src/mak/eval/classifier_eval.py - Classifier evaluation → Query parser: evaluates parser
src/mak/eval/classifier_eval.py