Machine Learning / AI

Как работал self-correction в Text-to-SQL агенте?

How was the inference of LLM (DeepSeek R1 33B) organized? How many GPUs, and how was parallelization handled?

Tell about the architecture of the RAG-chat (iChat) in TechVille: pipeline, components, technologies.

Tell me about the Text-to-SQL agent: architecture, pipeline, technologies.

What metrics were used to evaluate the RAG system and how was the decision made to deploy it to production?

How was the quality of the Text-to-SQL agent evaluated?

Tell us about the report summarization project at Alrosa: task, approach, fine-tuning.

Why did you choose Qdrant instead of staying with Elasticsearch? What are the advantages?

How were load peaks handled in the RAG system? What happened under high load?

What ranking metrics were used to evaluate the retriever?