Public profile
Alibi Serikbay
@speedruntheworld
Senior AI Engineer · BTS Digital
AI/ML Engineer with 5 years building production LLM systems — RAG pipelines, agentic workflows, and the evaluation harnesses that prove they actually work.
About
Recent: hybrid retrieval (pgvector HNSW + BM25, fused via RRF) that lifted nDCG 18.9%; a legal RAG system that moved macro recall from 64% → 81.6% across a 1,200-query DeepEval suite; a LangGraph ReAct agent routing vector search, structured APIs and web tools through a LiteLLM gateway; an MCP server exposing 10 tools to any MCP-compatible client. Before LLMs: computer vision — YOLO detection, ViT-based embedding search, liveness detection for anti-spoofing. Stack: Python · PyTorch · LangChain/LangGraph · FastAPI · PostgreSQL/pgvector · Docker · AWS · DeepEval · Langfuse
How I can help
open-source contribution
Looking to connect
jobs, research and papers