🌀 Convector – CSV → JSONL (384‑dim) + Qdrant Quickstart

Autor: Roger Camara

Convector is a tiny, practical toolkit to turn .csv datasets into newline‑delimited JSON (output.jsonl) with 384‑dim sentence embeddings and the original row as payload. It pairs with a simple importer to load the file into a local Qdrant vector DB (Docker).

✨ What you get

• convector.py – reads your CSV, auto‑detects columns, builds one text per row, generates 384‑dim embeddings, and writes output.jsonl:

  {"id":"<uuid>", "text":"<row-as-text>", "vector":[...384 floats...], "payload":{...original row...}}

• qdrantimport.py – asks for output.jsonl, lists Qdrant collections, and imports in batches with a progress bar.

We use a free embedding model (paraphrase-multilingual-MiniLM-L12-v2) that outputs 384 dimensions. If you switch to another provider (e.g., OpenAI), you can use larger vectors—just make sure your collection size matches.

📦 Requirements

• Python 3.9+

• Install deps:

  pip install -r requirements.txt

---

🚀 Convert your CSV → JSONL

1. Put your .csv next to convector.py.

2. Run:

   python convector.py

3. Paste/drag your CSV path and confirm the detected columns.

4. You’ll get output.jsonl in the current folder.

---

🧪 Qdrant (Docker) — ultra‑quick setup

1) Start Qdrant locally

docker run -p 6333:6333 --name qdrant --rm qdrant/qdrant

2) Create a collection with 384 dimensions

curl -X PUT "http://localhost:6333/collections/my_collection"   -H "Content-Type: application/json"   -d '{"vectors": {"size": 384, "distance": "Cosine"}}'

3) Import your JSONL

python qdrantimport.py

• Enter output.jsonl path.
• Press Enter to keep default Qdrant URL (http://localhost:6333).
• Select my_collection.
• Watch the progress bar until ✅ Done.

---

🔍 Quick query test

Search by vector

curl -X POST "http://localhost:6333/collections/my_collection/points/search"   -H "Content-Type: application/json"   -d '{
    "vector": [0.1, 0.2, ... 384 floats ...],
    "limit": 3
  }'

Search by payload filter

curl -X POST "http://localhost:6333/collections/my_collection/points/scroll"   -H "Content-Type: application/json"   -d '{
    "filter": {
      "must": [
        {"key": "payload.column_name", "match": {"value": "some_value"}}
      ]
    },
    "limit": 3
  }'

📝 Notes

• Output is always output.jsonl in the current folder.
• IDs are deterministic (UUID5) for reproducible imports.
• Free model = 384 dimensions. Change model & collection size together.