Quang Le

Quang Le

Senior Python Developer / Full Stack Engineer

I build back ends. Since March 2022 I have been a backend engineer at SDS Manager in Norway, where most of my work is the unglamorous kind that decides whether a product feels fast: query optimisation, search relevance, caching, and the logging you need to find out what actually broke.

Four years of Python — Django, Flask and FastAPI, REST and async endpoints, including moving MVC codebases to async architectures. Four years of JavaScript and TypeScript, more than three of them on React. More recently I have been integrating Google Gemini and the Anthropic Claude API into production workflows, and building MCP servers so those models have real tools to call.

What I work on

  • Backend performance. Schema design, N+1 query elimination, indexing and complex conditions across SQL Server, PostgreSQL and MongoDB; Redis and Memcached in front of the expensive paths.
  • Search and relevance. Elasticsearch, including custom scoring, plus the rest of the ELK stack for logging and monitoring.
  • Data extraction. Turning messy documents into structured records reliably enough to build features on.
  • LLM tooling. Gemini and Claude API integrations, and MCP servers for tool-augmented interactions.
  • Real-time and async. Celery, RabbitMQ, server-sent events, WebSockets and polling, on Linux and Docker.

Selected work

  • Making a slow product fast

    SDS Manager

    Rewrote the queries behind the heaviest pages for a 400% improvement in load speed, then moved the statistical and analytical APIs onto Redis, cutting database load by 60%.

  • Search that tolerates typos

    SDS Manager

    Customised Elasticsearch scoring so misspelled queries still find the right safety data sheet, and reworked relevance around how people actually search.

  • A document pipeline you can trust

    SDS Manager

    Hardened a complex safety data sheet extraction pipeline to roughly 95% accuracy, and added Kibana and Logstash monitoring that cut issue resolution time by 80%.

  • LLMs with real tools

    SDS Manager

    Integrated Gemini and the Claude API to automate internal workflows, and designed and implemented MCP servers so the models have proper tools to call rather than prompts to guess from.

More on what I have built →

Writing

All posts →

Elsewhere