Weekly review June 26th - July 2nd 2020

Currently I am taking some time between roles to study MLOps, AI product development, and furthering my study of System Design.

I am starting to interview for jobs. I am also looking for freelance and contract positions to keep my time and commitment more flexible.

I want to lead software product teams in climate tech and sustainability, and for this I want to be ready to jump into any exciting project.

Smart Buildings / IoT job

This week I got contacted on LinkedIn for a CTO or VP of Engineering role for a company working on IoT / Smart Buildings. It's a few small teams right now but they have Series A funding and are ready to expand. I'm focused on MLOps and architecture right now, but I've also found that technical management is often lacking and a lot of products die because they have non-technical managers.

So maybe I should do less (or no!) coding and again focus on people and debugging those problems in how we build products. I had my first meeting with the company this week.

But I still want to be focused on climate tech.

Building automation is on the drawdown list though: drawdown.org/solutions/building-automation-..

Re-reading the later chapters of The Manager's Path

Studying more architectural design patterns.

I should be: working more on architecture and org diagrams in Figma.

I really want to be working on climate and ESG data. Learning how AI products are developed; the whole process especially production.

Infinite Album

Did a zoom call with Ryan Groves from Infinite Album which was quite nice. They have a Twitch extension for streamers that hooks into the game and creates constant quality music.

He's looking for Python developers to work on his AI music startup. I told him it wasn't the direction I was going in but we agreed to talk because it would be interesting to connect.

I used to do a lot of music programming and I wanted to find out what he was doing and connect him with others I might know. We have to have coffee together soon in real life.

Evaluating libraries and tools

DVC

Track models and datasets.

The tool syncs the actual files to cloud (bucket etc) and commits a text file with MD5 hash to your project's git. This makes it easy to have big data files in your project source without blowing up your git (and especially your .git folder)

Feast

FeatureStore - a tool for working with datasets, publishing them to a registry for production usage other team members to use.

It connects to your Data Warehouse (local folder, remote bucket, SnowFlake, BigQuery, Redshift etc.)

I worked with this a fair bit this week.

The first FeatureStore was Uber's Michelangelo.

The commercial version of Feast is Tecton.

I like the usage pattern quite a bit. I will be using it for Sensilla.io shortly.

BERT interlinking tool

twitter.com/LeeFootSEO/status/1540021801455.. [[BERT]] interlinking Tool. I think it finds related pages to cross link to for SEO optimization. I'm interested in it for other purposes.

Uses [[Streamlit]] for the UI github.com/searchsolved/search-solved-publi..

Streamlit is hilarious. It's very playful; the code is for scripters and hackers, not for correctness. As usual Python gets abused by the Data Science crowd, but I love it. You can deploy usable interfaces for demos and testing super quick and it looks really good.

markwhen.com Markwhen markdown Gantt chart maker

github.com/mingrammer/diagrams Diagrams as code

DevOps and whatnot

twitter.com/danielbryantuk/status/153853010.. Platform engineering tips. Buildpacks, Telepresence (develop locally, hotload into an online Kubernetes pod), Skaffold (hotloading to local or remote Kubernetes) etc.

Divergence Convergence Pendulum

“Divergence” refers to opening up your senses and taking in new sources of information from the outside world, such as at the start of a new project. “Convergence” refers to shutting off sources of distraction and narrowing your focus to arrive at an end result.

trpc

trpc works with react-query, it gives us [[RPC]] for node + React. Works well with the lovely Zod.

sociomapping:

Sociomapping is a method developed for processing and visualization of relational data (e.g. social network data). It is most commonly used for mapping the social structure within small teams (10-25 people).

I like the idea of generating diagrams from data. Cliques and actionable mini networks, as a [[Sensilla data products]]

Art generated from networks but without the hairballs. Skew and time series, propagation.

Tweet NLP

tweetnlp.org

TweetNLP is a website to enable users to use cutting-edge language technologies in social media, irrespective of their level of expertise.

TweetNLP contains both demos and tutorials with many examples to get you started. Best of all, everything is free!

Tasks supported:

Sentiment analysis, emotion recognition, hate speech detection, offensive language identification, emoji prediction, topic classification, named entity recognition

Next JS Middleware

Next.js edge Middleware is released.

nextjs.org/docs/advanced-features/middleware

I love the idea of running these little edge functions and being able to activate feature flags or A/B testing with minimal code and have it be performant. I haven't tried these yet.

This is also a lovely diagram:

Pasted image 20220701100025.png