LLM gateways, guardrail latency and quality checks in CI

26 August 2026#

  • Vectorize. Cloudflare provides a vectordb solution as well, with a generous free tier. Would be good to add this into my website chatbot solution.

27 August 2026#

  • Lighthouse. Lighthouse is an open-source, automated tool to improve the quality of web pages. It has audits for performance, accessibility and SEO. Can be integrated with CI/CD before any new release.
  • Lychee. Lychee link checker is a good fit for content-driven websites. Broken links can be found and fixed at the CI level itself.

29 August 2026#

  • Guardrails. Implementing guardrails in AI applications at times significantly increases the app’s latency.
  • Prompt injection. SFT used to defend LLMs against prompt-injection attacks often fails by relying on surface-level patterns rather than understanding malicious intent. Good read: arxiv.org/pdf/2601.07185
  • LLM gateways. Explored tensorzero while exploring open-source LLM gateway options. Another option worth exploring for enterprise: Portkey gateway. TrueFoundry, Kong are some others.
  • Caching. Caching, when not implemented properly, can cause data leaks. Important to consider in the AI application world.

30 August 2026#

  • Web scraping. Scrapling is a web scraping framework with anti-bot and Cloudflare bypass capability.
  • Chatbot feedback. User feedback extracted from chatbot conversations can be used for evaluation and personalization of the application.