Fix Local Qdrant RAG Drift

I run an on-premise Retrieval-Augmented Generation stack in Riyadh that relies on Qdrant (latest stable) inside a Docker container. Over time I’ve seen clear retrieval regression and semantic drift: passages that once ranked high are now buried, and some queries pull semantically off-target results.

The job is fully on-site in Saudi Arabia—our servers are air-gapped—so I need someone already in the country who can sit with the team, dive into the containers, and restore relevance.

You will first trace the root cause of the drift (distance metrics, collection parameters, sharding, embedding pipeline or anything else at play). Once pinpointed, you’ll apply a durable fix within the existing Docker-compose setup and then verify the outcome through quantitative testing. The stack you’ll touch includes Python 3.11, LangChain, OpenAI embeddings and Qdrant 1.8.

Deliverables
1. Technical report that explains the diagnosed cause.
2. Updated Docker files, configs or migration scripts containing the remedy.
3. Test suite plus benchmark results proving retrieval quality is back to—or better than—its original baseline.

Acceptance criteria
• Recall and relevance scores match or exceed the April benchmark (figures provided on site).
• All containers build and run cleanly with no new regressions.

If you have hands-on Qdrant expertise, strong vector search optimisation skills and are available for immediate on-prem work, let’s get this fixed.