Professional work / 2022—present
Enterprise AI Platform
Brought together model routing, inference, RAG, enterprise knowledge and workflows into a platform for real operational tasks.
Context
The difficult part of enterprise AI is rarely the first model connection. Permissions, knowledge, tools, evaluation and iteration must work together. This case combines methods that can be discussed publicly across several internal platforms.
My role
- Owned product engineering and technical direction across AI and data teams.
- Evaluated commercial and open models and built inference services with FastChat and vLLM.
- Built RAG and enterprise knowledge workflows with LangChain, Faiss and document pipelines.
- Delivered report generation, operations, policy Q&A, code, SQL and data-processing workflows.
- Iterated on quality, permissions and process failures based on real use.
Key judgment
Platform value does not come from the number of assistants. It comes from whether a small set of frequent tasks is dependable. Models, retrieval, tool use and business permissions have to be designed as one system.
Disclosure
This page excludes internal interfaces, client data, active-user figures and commercial metrics. The architecture diagram is a privacy-safe reconstruction.