seed_supabase.py

import os
from datasets import load_dataset
from langchain.schema import Document
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import SupabaseVectorStore
from supabase import create_client

# 1. Load GAIA train split
dataset = load_dataset("gaia-benchmark/GAIA", split="train")

# 2. Build Documents: "Q: …\nA: …"
docs = []
for ex in dataset:
    q, a = ex["question"], ex["answer"]
    docs.append(Document(
        page_content=f"Q: {q}\nA: {a}",
        metadata={"task_id": ex.get("task_id"), "split": "train"}
    ))

# 3. Initialize embedding & Supabase client
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
supabase_url = os.environ["SUPABASE_URL"]
supabase_key = os.environ["SUPABASE_SERVICE_KEY"]
supabase = create_client(supabase_url, supabase_key)

# 4. Upload to Supabase
vectorstore = SupabaseVectorStore.from_documents(
    docs,
    embedding=embeddings,
    client=supabase,
    table_name="documents",
    query_name="match_documents_langchain"
)

print(f"Seeded {len(docs)} GAIA examples into Supabase.")