22.01.26 (inclusive)

By agreement

MVP

ICT tasks

Media sphere

Neurotechnology and artificial Intelligence

Software/ IS

There is no centralized and standardized way to receive embeddings in the system — clients (kb-retriever and future services) are either tied directly to Ollama and its contract, or use “stubs/heterogeneous implementations”, which makes it difficult to ensure reliability, scaling and replacement of the model/provider without rewriting the code.

embedding-service: • The service starts and responds to /liveness, /readiness • POST /v1/embeddings vectorizes the text correctly • The dimension of the vectors corresponds to the model • Error handling: timeout → 503, connection refused → 503, invalid JSON → 400 • Retry strategy is working (check logs) • Unit tests coverage> 80% • Integration tests are being conducted • Dockerfile with multi-stage build • README with instructions Integration with kb-retriever: • Client created in internal/services/embedding/client.go • Configuration updated • Vectorization stubs replaced with real calls • End-to-end test passes: creation → vectorization → search

Pavlovsky Dmitry Alexandrovich