Writing
100 Days of Gen AI: Day 34
Where I left off last, the issue was that the Apple shortcuts app worked, but only for an hour until the spotify session token expired. Today I focused on removing...
100 Days of Gen AI: Day 33
There’s a feature I’ve wanted in Siri for a long time. The use case is I’m listening to music while driving, or cooking in the kitchen, or doing some other...
100 Days of Gen AI: Day 32
I got the AI guardrail filter working, kinda. The web UI is connected to a python server which can interact with my guardrail policy in AWS. All the UI and...
100 Days of Gen AI: Day 31
I started work on a prototype for LLM prompt guardrails. It filters for PII and masks it before sending the query on to the model for inference. It can also...
Building an LLM-Powered Slackbot
Here’s how we built a Retrieval-Augmented Generation (RAG) LLM-powered Slackbot at Benchling. You can read the full article on the Benchling Engineering blog. Read the post...
100 Days of Gen AI: Day 30
I read lots about fine-tuning today. I started with the OpenAI Fine-Tuning documentation, then branched out into several blog posts and perused many datasets on Hugging Face. My goal...
100 Days of Gen AI: Day 29
Today I augmented my slackbot knowledge base with some additional data sources. In addition to the existing confluence data source, I also added relevant slack threads and public documentation from...
100 Days of Gen AI: Day 28
Today I focused on making this slackbot fully production-ready. I set up a permanent dev and prod environment (I’d been using a sandbox thus far). I also documented a runbook...
100 Days of Gen AI: Day 27
Today I did a few things: Verified incoming requests are actually from Slack Wrote out several unit tests Evaluated the quality of LLM responses I used ChatGPT 4o...
100 Days of Gen AI: Day 26
Today I started refactoring the slackbot infrastructure components into a reusable terraform module. This consists of an API Gateway, a Lambda (both using public modules), plus an IAM policy and...