The author reflects on the challenges faced during the development process, including working with rapidly deprecated Google APIs, dealing with timezones, and setting up LLMs to output structured data. Despite these challenges, the author believes that LLMs have incredible potential in automation and can add a lot more value if a working standard for structured output is established. The author concludes by sharing a link to the code for anyone interested in trying out the tool.
Key takeaways:
- The author and a friend developed a tool to help manage medium priority actionable emails by adding tasks directly to a calendar using a chatbot.
- The tool was initially built using closedAI’s ChatGPT for proof of concept, but was later replaced with open-source LLMs like Mistral due to security concerns.
- The author faced challenges with Google APIs, timezones, and structured output from LLMs during the development process.
- The author concludes that despite the challenges, LLMs have great potential in automation and can add significant value if a working standard for structured output can be established.