How ineshag/srmist_puchkaz_prism_genai works
The product is a guided device-troubleshooting API. A customer complaint enters through the API, is enriched into a canonical query and paraphrases, then checks the semantic cache. On a miss, the pipeline extracts a structured answer from relevant reference text, maps it to deeplinks, sanitizes it, and returns a validated response; eligible results can then be cached. README documents the stages and optional LLM providers. Some implementation details, especially API and deeplink-mapping internals, were not excerpted, so their links rely on the README or source-index references.
Main entry points: pipeline.py Pipeline orchestrator · api.py Troubleshoot API · structure_extraction.py Structure extraction · cache.py Semantic cache
Components 14
Request and response
- Troubleshoot API
src/api.py
Troubleshooting stages
- Query enrichment
src/enrichment.py - Pipeline orchestrator
src/pipeline.py - Structure extraction
src/structure_extraction.py - Relevance gate
src/relevance_gate.py - Deeplink mapping
src/deeplink_mapping.py - Response scrubber
src/scrubber.py
Runtime services
- Semantic cache
src/cache.py - TF-IDF embedder
src/embeddings.py - LLM client
src/llm_client.py
Response contract
- Response schema
src/schema.py - API response models
src/response_models.py
Other components
- Customer
- LLM providers
Connections 17
- Customer → Troubleshoot API: submits complaint
README.md - Troubleshoot API → Pipeline orchestrator: runs troubleshooting
src/api.py - Troubleshoot API → API response models: validates API response
src/api.py - Pipeline orchestrator → Query enrichment: enriches query
src/pipeline.py - Pipeline orchestrator → Semantic cache: checks cache
src/pipeline.py - Semantic cache → TF-IDF embedder: encodes query
src/cache.py - Pipeline orchestrator → Structure extraction: extracts structure
README.md - Structure extraction → Relevance gate: checks relevance
src/structure_extraction.py - Pipeline orchestrator → Deeplink mapping: maps deeplinks
README.md - Pipeline orchestrator → Response scrubber: sanitizes response
src/pipeline.py - Pipeline orchestrator → Response schema: validates response
src/pipeline.py - Structure extraction → Response schema: builds response
src/structure_extraction.py - Query enrichment → LLM client: requests paraphrases
README.md - Structure extraction → LLM client: requests extraction
src/structure_extraction.py - LLM client → LLM providers: calls provider
README.md - Pipeline orchestrator → Semantic cache: stores eligible result
src/pipeline.py - Troubleshoot API → Customer: returns response
README.md