Jiaming Nie

Founding Software Engineer specializing in applied AI, LLM training and evaluation, multi-agent systems, and production backend infrastructure.

Email · LinkedIn · GitHub

Last updated: July 2026.

Skills

• AI/ML: LLM fine-tuning (SFT and RFT), prompt engineering, RAG, multi-agent systems, model evaluation, and benchmarking
• Models and APIs: OpenAI (GPT-4.1 and o4-mini), Anthropic Claude, and Twilio Voice AI
• Languages: Python, JavaScript, and TypeScript
• Frameworks and infrastructure: FastAPI, React, MongoDB, Redis, Docker, AWS, Jenkins CI/CD, and Prometheus/Grafana

Experience

Youlify — Remote

Founding Software Engineer · January 2025–Present

• Built and operated OpenAI-hosted SFT and RFT workflows for medical billing code extraction. Curated office and surgery datasets covering more than 2,000 encounters, generated practice-stratified training and evaluation datasets from clinical notes and PDF-derived documentation, and launched GPT-4.1 SFT and o4-mini RFT jobs through the OpenAI Fine-Tuning API.
• Implemented the supporting evaluation stack, including token and cost tracking, checkpoint evaluation, benchmark comparisons, and a Python grader using weighted edit distance, CPT prefix decay, and ICD/modifier overlap.
• Built production AI workflows for healthcare revenue cycle management, including Charge Verifier and Denial Agent, with structured claim context, fine-tuned model routing, Redis-backed task state, and asynchronous appeal letter generation.
• Developed a Redis-backed WebSocket notification platform that gives users real-time visibility into patient and document sync, AI EOB parsing, charge verification, denial analysis, and policy monitoring.
• Implemented bidirectional WebSocket proxying for Twilio voice streams and server-sent event streaming for AI copilot conversations across portal and practice-management products.
• Helped evolve the platform from a monolithic backend into FastAPI services spanning practice management, portal, integration, authentication, and GenAI capabilities.
• Built and maintained Selenium and Playwright EHR automation for document download, patient and encounter sync, and operative-note retrieval across systems including Athena and Practice Fusion.

peerup — Shanghai, China

Software Engineer · June 2023–January 2025

• Led backend development for an AI assistant and chatbot in a note-taking product using Python and FastAPI.
• Built a RAG system with Milvus and Elasticsearch to retrieve from public, private, and third-party knowledge sources.
• Delivered content enrichment features including text segmentation, keyword highlighting, and public-reference retrieval.
• Served as a project architect, leading technology selection, system design, project planning, CI/CD, containerization, and engineering conventions.

Oqton — Shanghai, China

Software Engineer · October 2021–February 2023

• Developed Flask applications that communicated with industrial machine APIs, collected telemetry, and issued commands.
• Deployed and maintained the applications in Docker on Linux gateways.
• Investigated production issues using Elasticsearch and Apache Kafka.
• Built Node.js and Express APIs plus React features for machine configuration, filtering, and sorting.

Micro Focus — Shanghai, China

Software Engineer · July 2020–September 2021

• Developed web-component detection features using JavaScript, C++, and object recognition.
• Implemented cross-domain element detection for web automation in Node.js.
• Improved the customer-case workflow, reducing average development-stage handling time by four days per case.

Selected Projects

Medical Billing Code Extraction with LLM Fine-Tuning

February 2026–Present

• Fine-tuned GPT-4.1 with SFT and o4-mini with RFT on more than 2,000 clinical encounter-to-billing-code pairs across office and surgery domains.
• Designed an RFT reward function using weighted edit distance, CPT prefix decay, ICD/modifier Jaccard distance, and exponential scoring.
• Built the pipeline from clinical-note CSV and PDF extraction through CPT, ICD-10, and modifier normalization to SFT/RFT JSONL generation, token counting, and cost estimation.
• Evaluated CPT, ICD-10, modifier, and full-code accuracy with macro/micro precision, recall, and F1 across checkpoints and base models.

Real-Time Infrastructure for AI Agent Workflows

January 2025–Present

• Designed a WebSocket notification layer for multi-stage AI workflows, replacing REST polling with real-time task progress.
• Built a Redis Cluster-backed asynchronous task state machine with sub-state tracking, notification deduplication, and reliable final-state delivery.
• Implemented server-sent event streaming for AI copilots while persisting complete messages to MongoDB and triggering conversation summarization.

AI Assistant and Chatbot for a Note-Taking Product

September 2023–January 2025

• Developed a FastAPI backend for an AI assistant with streaming responses, instruction execution, and file-based question answering.
• Built hybrid RAG with Milvus and Elasticsearch across public, private, and web sources.
• Implemented agent workflows with LangChain and LlamaIndex and Redis-backed conversation, message, attachment, and user-context management.

Education

Worcester Polytechnic Institute — Massachusetts, United States

Master of Science, Computer Science · August 2017–December 2019

University of Liverpool — Liverpool, United Kingdom

Bachelor of Engineering, Electrical Engineering · August 2015–June 2017

Xi’an Jiaotong-Liverpool University — Suzhou, China

Bachelor of Engineering, Electrical Engineering · August 2013–August 2015