Day 8: Streamlit UI + dotenv config

.gitignore

@@ -4,3 +4,4 @@ data/issues/
 data/chroma/
 __pycache__/
 .pytest_cache/
+.env

app.py

@@ -0,0 +1,111 @@
+"""Streamlit UI for the FastAPI Codebase Q&A RAG system.
+
+Run locally:
+    export GROQ_API_KEY=gsk_...
+    streamlit run app.py
+
+On Hugging Face Spaces, set GROQ_API_KEY as a Space secret.
+"""
+
+import os
+import sys
+from dotenv import load_dotenv
+load_dotenv(override=True)
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).parent / "src"))
+
+import streamlit as st
+
+from ask import SYSTEM_PROMPT, build_prompt
+from retrieval import retrieve
+
+LLM_MODEL = os.environ.get("GROQ_MODEL", "openai/gpt-oss-120b")
+
+EXAMPLES = [
+    "How do I return a custom status code from an endpoint?",
+    "How does dependency injection work?",
+    "Where is the APIRouter class defined?",
+    "How do I connect FastAPI to MongoDB?",  # tests honest refusal
+]
+
+st.set_page_config(page_title="FastAPI Codebase Q&A", page_icon="⚡",
+                   layout="wide")
+st.title("⚡ FastAPI Codebase Q&A")
+st.caption("Retrieval-augmented answers over FastAPI's source code, docs, "
+           "and GitHub issues — every claim cited, honest refusals when "
+           "the corpus doesn't know.")
+
+with st.sidebar:
+    st.header("How it works")
+    st.markdown(
+        "1. Your question is embedded (`bge-small-en-v1.5`)\n"
+        "2. Top-5 chunks retrieved from ~1,350 AST-aware chunks "
+        "(code, docs, issues)\n"
+        "3. An LLM answers **only** from those chunks\n"
+        "4. Citations link to the exact GitHub lines"
+    )
+    mode = st.selectbox("Retrieval mode", ["dense", "dense_rw", "hybrid"],
+                        help="dense won the ablation; others shown for "
+                             "comparison")
+    st.divider()
+    st.markdown(
+        "**Eval results (42-question set)**\n\n"
+        "recall@5 **0.91** · MRR **0.71**\n\n"
+        "faithful **0.89** · correct **0.91** · refusal **7/7**"
+    )
+    # TODO: replace with your repo URL
+    st.markdown("[Source & write-up](https://github.com/YOUR_USERNAME/codebase-rag)")
+
+
+@st.cache_data(show_spinner=False)
+def cached_retrieve(question: str, mode: str) -> list[dict]:
+    return retrieve(question, k=5, mode=mode)
+
+
+def generate(question: str, hits: list[dict]) -> str:
+    from groq import Groq
+    api_key = os.environ.get("GROQ_API_KEY")
+    if not api_key:
+        st.error("GROQ_API_KEY is not set.")
+        st.stop()
+    client = Groq(api_key=api_key)
+    response = client.chat.completions.create(
+        model=LLM_MODEL,
+        messages=[
+            {"role": "system", "content": SYSTEM_PROMPT},
+            {"role": "user", "content": build_prompt(question, hits)},
+        ],
+        temperature=0.1,
+    )
+    return response.choices[0].message.content
+
+
+# --- example question buttons ---
+st.write("Try one:")
+cols = st.columns(len(EXAMPLES))
+for col, ex in zip(cols, EXAMPLES):
+    if col.button(ex, use_container_width=True):
+        st.session_state["question"] = ex
+
+question = st.text_input("Ask about FastAPI's codebase",
+                         key="question",
+                         placeholder="How do I handle file uploads?")
+
+if question:
+    with st.spinner("Searching the codebase..."):
+        hits = cached_retrieve(question, mode)
+    with st.spinner("Writing the answer..."):
+        answer = generate(question, hits)
+
+    st.markdown(answer)
+
+    st.divider()
+    st.subheader("Sources")
+    for i, h in enumerate(hits, 1):
+        meta = h["meta"]
+        label = (f"[{i}] {meta['source_type']} · {meta['path']}"
+                 + (f" · {meta['symbol']}" if meta["symbol"] else ""))
+        with st.expander(label):
+            st.markdown(f"[Open on GitHub]({meta['url']})")
+            st.code(h["text"][:1500])

requirements.txt

@@ -0,0 +1,5 @@
+streamlit
+chromadb
+sentence-transformers
+groq
+rank-bm25

src/ask.py

@@ -18,8 +18,8 @@ import os
 import sys
 
 from retrieval import retrieve as retrieve_chunks
-
-REFUSAL_TEXT = "I couldn't find this in the indexed codebase."
+from dotenv import load_dotenv
+load_dotenv(override=True)
 LLM_MODEL = os.environ.get("GROQ_MODEL", "openai/gpt-oss-120b")
 TOP_K = 5
 
@@ -74,10 +74,6 @@ def main() -> None:
     args = parser.parse_args()
 
     hits = retrieve_chunks(args.question, k=args.k, mode=args.mode)
-    top_score = hits[0].get("score")
-    if top_score is not None and top_score < SCORE_THRESHOLD:
-        print(f"\n{REFUSAL_TEXT}")
-        return
     
     if args.show_chunks:
         for i, h in enumerate(hits, 1):

src/eval.py

@@ -23,6 +23,8 @@ import time
 from pathlib import Path
 
 from retrieval import retrieve
+from dotenv import load_dotenv
+load_dotenv(override=True)
 
 DATA_DIR = Path("data")
 EVAL_SET = DATA_DIR / "eval_set.jsonl"