rag value updates

Co-authored-by: Copilot <copilot@github.com>

.env.example

@@ -11,3 +11,6 @@ RAW_TEXT_DIR=data/raw_text
 PDF_TIMEOUT_SEC=60
 PDF_MAX_RETRIES=3
 PDF_RETRY_BACKOFF_SEC=1.5
+RAG_CHUNK_SIZE=1400
+RAG_CHUNK_OVERLAP=250
+RAG_SYLLABUS_TOP_K=8

.gitignore

@@ -45,4 +45,5 @@ build/
 prompt.md
 stud_info.md
 test_db_to_api.py
-test.ipynb
+test.ipynb
+ingest_syllabus_to_neon.py

app/config.py

@@ -32,6 +32,9 @@ class Settings:
     pdf_timeout_sec: int = int(os.getenv("PDF_TIMEOUT_SEC", "60"))
     pdf_max_retries: int = int(os.getenv("PDF_MAX_RETRIES", "3"))
     pdf_retry_backoff_sec: float = float(os.getenv("PDF_RETRY_BACKOFF_SEC", "1.5"))
+    rag_chunk_size: int = int(os.getenv("RAG_CHUNK_SIZE", "1400"))
+    rag_chunk_overlap: int = int(os.getenv("RAG_CHUNK_OVERLAP", "250"))
+    rag_syllabus_top_k: int = int(os.getenv("RAG_SYLLABUS_TOP_K", "8"))
 
 
 settings = Settings()

app/main.py

@@ -84,7 +84,11 @@ def process_syllabus(courses: List[CourseInput]) -> SyllabusProcessResponse:
             if not syllabus_text:
                 raise RuntimeError("No text extracted from PDF.")
 
-            chunks = chunk_text(syllabus_text)
+            chunks = chunk_text(
+                syllabus_text,
+                chunk_size=settings.rag_chunk_size,
+                overlap=settings.rag_chunk_overlap,
+            )
             if not chunks:
                 raise RuntimeError("Unable to create text chunks from syllabus content.")
 
@@ -175,8 +179,12 @@ Rules:
                 neon_connect_timeout_sec=settings.neon_connect_timeout_sec,
             )
             query_embedding = gemini.embed_text(req.query, task_type="retrieval_query")
-            hits = vector_store.search(req.semester, query_embedding, top_k=5)
-            hits = hits[:5]
+            hits = vector_store.search(
+                req.semester,
+                query_embedding,
+                top_k=settings.rag_syllabus_top_k,
+            )
+            hits = hits[: settings.rag_syllabus_top_k]
             chunks_passed = len(hits)
             syllabus_context = "\n\n---\n\n".join(
                 [f"[{h.get('course_code', '')}] {h.get('chunk', '')}" for h in hits]
@@ -229,6 +237,9 @@ Rules:
         f"Language: {req.lang_code}\n"
         f"Student endpoint intent param: {intent if intent in {'attendance', 'result'} else 'not_called'}\n"
         f"Student endpoint semester param: {req.semester if intent in {'attendance', 'result'} else 'not_called'}\n"
+        f"Syllabus top_k configured: {settings.rag_syllabus_top_k}\n"
+        f"RAG chunk size configured: {settings.rag_chunk_size}\n"
+        f"RAG chunk overlap configured: {settings.rag_chunk_overlap}\n"
         f"Syllabus chunks passed: {chunks_passed}\n"
         "LLM Called: yes\n"
         "\n--- Prompt Passed To LLM ---\n"