Unstructured Data Processing for AI

from mixpeek import Mixpeek

client = Mixpeek(api_key="YOUR_API_KEY")

# 1. Create a namespace (your data container)
namespace = client.namespaces.create(
    name="enterprise-docs",
    embedding_model="mixpeek-embed-v2"
)

# 2. Connect your object storage
bucket = client.buckets.create(
    namespace_id=namespace.id,
    source="s3://company-data/unstructured/",
    credentials={"role_arn": "arn:aws:iam::role/mixpeek"}
)

# 3. Define a processing collection
collection = client.collections.create(
    namespace_id=namespace.id,
    bucket_id=bucket.id,
    extractors=[
        {"type": "text_embedding"},
        {"type": "image_embedding"},
        {"type": "video_keyframe"},
        {"type": "audio_transcription"},
        {"type": "ocr"},
        {"type": "object_detection"}
    ]
)

# 4. Files are processed automatically when they land in the bucket.
#    Query across all modalities with a single retriever:
results = client.retrievers.execute(
    namespace_id=namespace.id,
    stages=[
        {
            "type": "feature_search",
            "method": "hybrid",
            "query": {"text": "quarterly revenue projections"},
            "limit": 20
        },
        {"type": "rerank", "model": "cross-encoder", "limit": 5}
    ]
)

# Results span PDFs, slide decks, video recordings, and images
for r in results:
    print(f"{r.modality}: {r.content[:100]}  (score: {r.score})")