At Arbisoft, I design and deploy production ML systems using pretrained models, prompt engineering, and backend components. I fine-tuned Qwen3.5-4B for CV parsing and information extraction to support downstream ranking and matching.
I built a production RAG-powered library search agent with LangGraph and LangChain, using semantic search, metadata filtering, and LLM reasoning to find books in a digital library. I also designed an agent-based pipeline that generates structured, illustrated stories, with QA agents to detect hallucinations, logical errors, and formatting issues.
For distributed graph analytics, I migrated a knowledge-graph workload from single-instance Neo4j to NebulaGraph and re-implemented analytics with cuDF and Dask-cuDF. I scaled the pipeline to a multi-machine, multi-GPU cluster and built monitoring dashboards with Grafana and Prometheus.
Earlier, at Naeco Blue, I trained and evaluated LSTM and other machine-learning models for renewable energy forecasting, working with weather APIs to obtain and preprocess data. I also built a privacy-preserving tool that detects and redacts student PII from scanned homework submissions.
