Rakhat Tuknazarov

Designed and experimented with LLM-based agent workflows using LangChain (prompt templates, tool-calling, structured outputs). Implemented early-stage agent orchestration patterns (multi-step reasoning, tool-use, fallback logic) for internal automation use-cases. Vision & OCR Research: Evaluated and integrated various OCR engines (Tesseract, PaddleOCR, or others you used) to enable agents to process complex, multi-modal documents.

Configured and managed GPU-accelerated environments on RunPod for LLM inference. Optimized model serving using vLLM, significantly reducing latency and operational costs. Designed and implemented end-to-end LLM-powered automation for compliance and risk assessment workflows (requirements architecture deployment). Integrated LLM services with existing backend systems via REST APIs and async task queues. Defined basic evaluation metrics (latency, response consistency, failure cases) for internal AI-driven services. Built and optimized specialized Docker images for LLM workloads, ensuring seamless deployment across cloud and on-premise environments.