rameshucs06/subtitle_rag

How rameshucs06/subtitle_rag works

Bilingual Subtitle Search & Q&A is a Streamlit front end over a FastAPI service for English/Telugu subtitle search, cited answers, episode summaries, and monitoring. The main path is user → UI → API → retrieval and (for answers) RAG generation; Telugu questions are translated for English retrieval. Cleaning and chunking, evaluation, and monitoring are distinct supporting capabilities. The README documents the index and product capabilities; links not shown in the excerpts or source references are omitted rather than inferred.

Main entry points: main.py HTTP API · search.py Hybrid retriever · rag.py RAG workflow · llm.py Gemini wrapper

Components 24

User experience

Ask UI ui/app.py
Inspect page ui/pages/1_Inspect.py
Summarize page ui/pages/2_Summarize.py
Monitor page ui/pages/3_Monitor.py
HTTP client ui/api_client.py

API orchestration

HTTP API api/main.py
Configuration src/subtitle_rag/config.py

Search and corpus

Hybrid retriever src/subtitle_rag/retrieval/search.py
BM25 index src/subtitle_rag/indexing/bm25.py
Chroma index artifacts
Subtitle chunks data/processed
Subtitle cleaning src/subtitle_rag/ingestion/cleaning.py
Chunk creation src/subtitle_rag/ingestion/chunking.py

Answer generation

RAG workflow src/subtitle_rag/generation/rag.py
Prompt templates src/subtitle_rag/generation/prompts.py
Episode summarizer src/subtitle_rag/generation/summarizer.py
Gemini wrapper src/subtitle_rag/generation/llm.py

Monitoring and evaluation

Request monitoring src/subtitle_rag/monitoring.py
Request log logs
Retrieval metrics src/subtitle_rag/evaluation/metrics.py
Answer evaluation src/subtitle_rag/evaluation/answer_eval.py
Evaluation questions data/eval/eval_questions.csv

Other components

User
Gemini service

Connections 26

User → Ask UI: asks or browses README.md
Ask UI → HTTP client: calls ui/app.py
Inspect page → HTTP client: calls ui/pages/1_Inspect.py
Summarize page → HTTP client: calls ui/pages/2_Summarize.py
Monitor page → HTTP client: calls ui/pages/3_Monitor.py
HTTP client → HTTP API: sends HTTP requests ui/api_client.py
HTTP API → Hybrid retriever: searches api/main.py
HTTP API → RAG workflow: asks api/main.py
HTTP API → Episode summarizer: summarizes api/main.py
HTTP API → Request monitoring: records and reads events api/main.py
Hybrid retriever → BM25 index: scores keywords src/subtitle_rag/retrieval/search.py
Hybrid retriever → Chroma index: queries and fetches README.md
Hybrid retriever → Subtitle chunks: loads chunk data README.md
RAG workflow → Hybrid retriever: retrieves passages src/subtitle_rag/generation/rag.py
RAG workflow → Prompt templates: builds prompts src/subtitle_rag/generation/rag.py
RAG workflow → Gemini wrapper: generates and translates src/subtitle_rag/generation/rag.py
Episode summarizer → Prompt templates: builds summary prompts src/subtitle_rag/generation/summarizer.py
Episode summarizer → Gemini wrapper: requests summaries src/subtitle_rag/generation/summarizer.py
Gemini wrapper → Gemini service: calls src/subtitle_rag/generation/llm.py
Gemini wrapper → Request monitoring: updates counters src/subtitle_rag/generation/llm.py
Request monitoring → Request log: writes and reads src/subtitle_rag/monitoring.py
Monitor page → HTTP API: requests metrics README.md
Answer evaluation → Retrieval metrics: uses hit ranking src/subtitle_rag/evaluation/answer_eval.py
Answer evaluation → RAG workflow: checks citations src/subtitle_rag/evaluation/answer_eval.py
Answer evaluation → Gemini wrapper: judges answers src/subtitle_rag/evaluation/answer_eval.py
Evaluation questions → Answer evaluation: evaluates against

Drawn by GitDiagram on November 14, 2026 from the repository's file tree, README and source files. Read as Markdown ·