Эрнис Бадмаев

I build reliable AI systems for document-heavy and expert workflows, and I care about ownership, traceability, and failure handling. My background combines Python backend development, applied ML, and production LLM/RAG systems, delivered end-to-end—from rules and data sources to APIs, retrieval, agent tools, evaluation, and operations.

In my current role as an AI/ML Engineer and AI Track Lead, I shipped an initial expert assignment pipeline that parses case materials, finds similar sections, ranks candidate experts, and produces audit logs. I also built an agent interface that shows the sources behind each recommendation and prepares drafts for manager approval, plus heterogeneous evidence search over XML, PDF, DOCX, and scans with file/page/fragment linkage. To keep AI from becoming hidden risk, I introduced scenario release gates with agent and document checks, ADRs on boundaries, and observability using Langfuse, Prometheus, and Grafana.

Previously, I developed churn prediction and early detection of troubled projects using CatBoost, and I used SHAP to explain decisions for managers. I also fine-tuned ruBERT for project-risk classification to quickly separate technical, resource, and organizational risks, supporting faster and more consistent owner assignment.

On a contract at Alice Tech (YC W25), I built an AI English tutor for a Telegram Mini App using an agent with memory, dialogue routing, and a voice pipeline (Whisper and ElevenLabs). I strengthened RAG with hybrid retrieval and reranking, and I created personalized exercise recommendations using a Neo4j graph and an event-driven pipeline.