Happy Friday (and Halloween weekend)!
We start this week with a quick project update. As mentioned in our last two editions, we’ve been looking to deploy Unsearch into some professional Slack communities for tech. We now have 2 confirmed pilots and 4 more demos in the works (covering over 15K+ members combined!) We’re not there yet obviously, but exciting news for our small team nonetheless!
Thanks to all of you who made community recommendations to us, they’ve been super helpful! And if you know of any other Slack communities looking to level-up, we’ll love to chat with them. Also, the tool’s 100% 🆓 so what’s not to like?
Featured Q&As
What are best practices in the QA testing process? We recently set-up two beta testing environments at my company. Wondering if that is overkill?
- Senior PM at E-commerce company
From Siaw Young, an experienced senior engineer previously at Forge and Carousell,
Would suggest looking at feature flags if your team isn't already using them in some shape or form. They are used to control rollouts or conduct experiments, but can also be used for QA to great effect
Keep in mind engineering and cognitive costs to maintaining many environments — need to generate realistic test data across all the environments, multiple copies of backend services, making sure the configuration is correct etc
Would also ask how the two beta testing environments will differ
How to improve battery life of consumer lifestyle product? We create adult toys to enrich people’s sexual wellness. We’re constantly trying to balance between the product being lightweight while still having long battery life. Any advice?
- Co-Founder, Early Stage, Health and Wellness
From John O’Brien, a Director at Chromatic and formerly at Butterfly Network,
Important to figure out if it’s an issue of active battery life, standby battery life or user expectations
It’s active battery life if the device doesn’t last long enough even after a fresh charge. If this is the case, you need to decrease power consumption or increase battery life. Note that you need to design to satisfy majority of users, not just the average one (no product can survive disappointing 50% of users)
It’s a standby battery issue if the charge doesn’t hold long enough when the user wants to use it next. Would start by reviewing the complexity of device (e.g., Bluetooth?) and using a power monitor from Monsoon solutions to measure standby power consumption rail by rail
User expectations is often the most overlooked. Without knowing too much about the product, I would suggest trying to make charging part of the user routine. Another option is to do some type of charging case like Airpods.
How to maintain your AI / ML models to avoid excessive data drift? We're a proptech start-up that built a simple AI/ML model to predict the market value of a property given a feed of past transactions in the area.
- Co-Founder, Early Stage, Prop Tech
From Hardik Gupta, an AI/ML data engineer at Arthur
Modern ML observability tools fix this by monitoring for data drift in the background and alerting the model owners when data drift actually occurs
Historically, data scientists have relied on retraining their models on a regular basis (daily, weekly, etc.) to ensure the model has the most fresh training data. This is naive and could result in wasting valuable compute resources or missing data drift if it happens suddenly
Worth checking out
Are you looking for a personal CRM tool to be more deliberate about your relationships? Check out Clay that was featured on Product Hunt earlier this week
Google vs GPT-3? See what Product Executive and serial entrepreneur David Weekly’s side by side comparison
Till next time!