Denis Kazhaev

AI Engineer, AI automation for sales and operational processes

1-lab

Built production AI agents and automation for sales, calls, outreach, and operations, including call quality monitoring across 25,950+ calls.

• Built a two-channel AI sales agent for a B2B equipment manufacturer. The agent on Avito collects a contact and moves the conversation to Telegram, keeping shared customer memory across channels. It answers using a RAG database of hundreds of SKUs in Google Drive, updated every 15 minutes, qualifies leads with a state machine, hands them to a manager using /connect and /disconnect, and records them in CRM. Runs 24/7 without losing context between channels.

ML Engineer

YADRO

Russian technology company that develops servers, storage systems, and other IT infrastructure products for business and enterprise customers.

Improved a technical documentation RAG assistant, increasing Recall@5 by about 18% and reducing unsupported claims by about 30%.

• Built a RAG assistant for technical documentation using hybrid search with dense retrieval and BM25, cross-encoder reranking, and product and version filters. Increased Recall@5 by about 18%. • Adapted a model using LoRA/PEFT and evaluated individual claims. Reduced the share of unsupported claims by about 30%. • Added caching and stage-by-stage optimization, reaching about 700 ms P50 latency. Used ClearML, Docker, Prometheus, and Grafana.

ML Engineer

ALP GROUP

Built a transcript-to-meeting-minutes system, increasing Recall@10 by about 12% and reducing hallucinations by about 20%.

• Built meeting minutes generation from transcripts using a classical ML baseline with scikit-learn, hybrid RAG with FAISS, BM25, and a reranker, hierarchical summarization, and Pydantic validation. • Increased Recall@10 by about 12% and reduced hallucinations by about 20%. Used FastAPI, Docker, GitHub Actions, and AWS.