AhmedShahan's picture
feat: implement Google Sheets writer module and add service account to ignore list
3215ea1
Raw History Blame Contribute Delete
7.55 kB
import os
from datetime import datetime, timezone
from dotenv import load_dotenv
from fastapi import FastAPI,HTTPException
from pydantic import BaseModel
from sentence_transformers import SentenceTransformer
from qdrant_client import QdrantClient
from groq import Groq
from groq import RateLimitError
from tavily import TavilyClient
import sheet_writer
load_dotenv()
# ── Config ────────────────────────────────────────────────────────────────────
COLLECTION_NAME = "portfolio"
TOP_K = 5
MODEL_NAME = "nomic-ai/nomic-embed-text-v1.5"
GROQ_MODEL = "llama-3.3-70b-versatile"
AI_NEWS_COUNT = 10
# ── Init ──────────────────────────────────────────────────────────────────────
app = FastAPI()
print("πŸ”§ Loading embedding model...")
embedder = SentenceTransformer(MODEL_NAME, trust_remote_code=True)
qdrant = QdrantClient(
url=os.getenv("QDRANT_URL"),
api_key=os.getenv("QDRANT_API_KEY"),
)
groq_client = Groq(api_key=os.getenv("GROQ_API_KEY"))
tavily_client = TavilyClient(api_key=os.getenv("TAVILY_API_KEY"))
# ── Schema ────────────────────────────────────────────────────────────────────
class QueryRequest(BaseModel):
question: str
# ── Routes ───────────────────────────────────────────────────────────────────
@app.get("/")
def root():
return {"status": "ok", "message": "Portfolio RAG API is running"}
@app.post("/query")
def query(request: QueryRequest):
# Step 1: Embed query
vector = embedder.encode(request.question, normalize_embeddings=True).tolist()
# Step 2: Search Qdrant
results = qdrant.query_points(
collection_name=COLLECTION_NAME,
query=vector,
limit=TOP_K,
).points
# Step 3: Build context
context = "\n\n---\n\n".join([r.payload.get("text", "") for r in results])
# Step 4: Ask Groq
prompt = f"""You are an AI assistant for Ahmed Shahan's portfolio.
Answer the user's question using ONLY the context below.
If the answer is not in the context, say "I don't have that information."
Context:
{context}
Question: {request.question}
"""
try:
response = groq_client.chat.completions.create(
model=GROQ_MODEL,
messages=[{"role": "user", "content": prompt}],
temperature=0.2,
)
return {
"question": request.question,
"answer": response.choices[0].message.content,
}
except RateLimitError as e:
# Extract retry time from error message
import re
match = re.search(r"try again in (.+?)\.", str(e))
retry_after = match.group(1) if match else "some time"
raise HTTPException(
status_code=429,
detail=f"Daily chat limit hit. Please try again after {retry_after}."
)
# ── AI News Endpoint ──────────────────────────────────────────────────────────
@app.get("/ai-news")
def get_ai_news():
"""Fetch today's top AI news, ranked by popularity, with simple explanations."""
try:
# Step 1: Fetch AI news from Tavily
search_result = tavily_client.search(
query="artificial intelligence AI news today",
topic="news",
time_range="day",
max_results=AI_NEWS_COUNT,
)
articles = search_result.get("results", [])
if not articles:
return {"articles": [], "message": "No AI news found today."}
# Step 2: Sort by popularity (score descending)
articles.sort(key=lambda a: a.get("score", 0), reverse=True)
# Step 3: Build a single prompt for Groq to simplify all articles
news_block = "\n\n".join(
f"--- Article {i+1} ---\n"
f"Title: {a['title']}\n"
f"Content: {a.get('content', 'No summary available.')}"
for i, a in enumerate(articles)
)
simplify_prompt = f"""You are a helpful assistant that explains AI news in simple, easy-to-understand language for a general audience.
Below are {len(articles)} AI news articles from today. For EACH article, provide:
1. The original title
2. A point-by-point explanation using VERY SIMPLE words that anyone can understand (like you're explaining to your grandmother)
Format your response EXACTLY like this (separate each article with a blank line):
===== Article 1 =====
Title: <original title>
Key points in simple words:
β€’ <point 1>
β€’ <point 2>
β€’ <point 3>
===== Article 2 =====
Title: <original title>
Key points in simple words:
β€’ <point 1>
β€’ <point 2>
...and so on for all articles.
Rules:
- Use ONLY simple, everyday words. No jargon.
- Each article should have 2-4 bullet points.
- Keep each bullet point short (1-2 sentences max).
- Focus on WHAT happened and WHY it matters.
Here are the articles:
{news_block}
"""
response = groq_client.chat.completions.create(
model=GROQ_MODEL,
messages=[{"role": "user", "content": simplify_prompt}],
temperature=0.3,
)
simplified_text = response.choices[0].message.content
# Step 4: Parse the response back into structured data
# Each article block starts with ===== Article N =====
import re
blocks = re.split(r"=====\s*Article\s+\d+\s*=====", simplified_text)
blocks = [b.strip() for b in blocks if b.strip()]
retrieved_at = datetime.now(timezone.utc).isoformat()
result_articles = []
for i, article in enumerate(articles):
explanation = blocks[i] if i < len(blocks) else "Explanation not available."
# Clean up the explanation - remove the "Title: ..." line since we have it separately
explanation_lines = explanation.split("\n")
# Filter out the "Title:" line if present
clean_lines = [
line for line in explanation_lines
if not line.strip().lower().startswith("title:")
]
clean_explanation = "\n".join(clean_lines).strip()
result_articles.append({
"title": article["title"],
"url": article["url"],
"published_date": article.get("published_date", ""),
"popularity_score": round(article.get("score", 0), 4),
"explanation_simple": clean_explanation,
})
# Write to Google Sheet (non-blocking β€” failures are logged, never crash the API)
sheet_writer.append_ai_news_articles(retrieved_at, result_articles)
return {
"data_retrieved_at": retrieved_at,
"timezone": "UTC",
"articles": result_articles,
"total": len(result_articles),
}
except Exception as e:
raise HTTPException(status_code=500, detail=f"Failed to fetch AI news: {str(e)}")
if __name__ == "__main__":
import uvicorn
uvicorn.run("api:app", host="0.0.0.0", port=8000, reload=True)