How did you work with queries in the vector storage (Qdrant) and their impact on the agent's latency?
Middle+
166
Machine Learning / AI
3611
0
What was the object of the state (TypedDict or other)?
Middle+
163
How did these four agents write to the shared state so as not to overwrite each other's state?
Middle+
161
You mentioned that the tools are read-only — was there any writing in the system?
Middle+
160
Did you consider the risk of prompt injection in the ReAct agent and how did you address it?
Middle+
160
Why did you use a multi-agent system with four agents instead of a single agent with tools (similar to ReAct)? What are the advantages of such a division?
Middle+
159
What is LoRA and why is it needed?
Middle+
158
How was the quality of generation evaluated (not retrieval)?
Middle+
157
How to properly split data into train, validation, and test sets for document NER?
Middle+
157
If one of the agents made a mistake, did the entire graph fail or reach the summary agent for degradation?
Middle+
157
Is it possible to build the retrieval part entirely on a cross-encoder, without a bi-encoder?
Middle+
156
What hyperparameters of LoRA do you know?
Middle+
155
Did you measure tokens and latency per pass to estimate the solution's cost?
Middle+
154
How was the reduction in information search time measured (metric in minutes)?
Middle+
152
What is LayoutLM and how does it differ from BERT for documents?
Middle+
151
Tell us about your approach: how do you build architecture with AI, how do you validate code, and how do you write tests?
Middle+
150
What do you record in the metadata of documents in a vector storage?
Middle+
150
Tell us about yourself and your relevant technical experience (banking domain projects: ML models, anti-fraud, RAG, multi-agent systems).
Middle+
149
Which ranker did you choose for reranking?
Middle+
149
Why did you use native toolcalling instead of passing a JSON schema in the prompt?
Middle+
149