We De-mystify Machine Learning

& Artificial Intelligence

Machine Learning and Artificial Intelligence can seem ​daunting, but they don’t have to be.

We can show you how to gain business insights and ​learn more about your customers using the data you ​already have today.

Neural Icon
Geometric Gradient Diagonal Line

Skip the Hype and Avoid ​the Snakeoil

The recent boom in ML/AI interest is one of the best and worst things to come along ​in a while for people like us: seasoned practitioners who were doing this ‘before it ​was cool’, when ML/AI was largely relegated to the realms of megacorp R&D and ​science labs.

We love this change because people are finally excited about stuff we’ve been ​working on for quite a while. But, this also means now there are are a lot of ​misconceptions out there about these technologies, too.

The terms ‘machine learning’ and ‘artificial intelligence’

are clever buzzwords that are an umbrella

for a bunch of probability math.

It’s not as mysterious as you think.

Case Study #1: The Internal Documentation Makeover

The Challenge

A large company wanted to minimize the amount of manual ​review needed to reorganize their stale internal ​documentation, as teams were having a difficult time finding ​the information they needed and much of it was out of date.

Our Approach

User Behavior Analysis: We looked at access patterns of ​the existing documentation to compare actual human ​behavior with self-reported preferences. We identified ​patterns we could use to tailor the documentation to match ​the user behavior better.

Clustering: Clustering groups data points (including text ​documents) into distinct clusters based on similarity, ​typically assigning each document to a single cluster ​without considering multiple topics. This showed us some ​emergent possibilities that we had not considered for top-​level topics in our information hierarchy.

Topic Modeling: This technique identifies and assigns ​multiple topics to documents based on their content, ​allowing for nuanced categorization of text. This helped us ​get on top of many of the larger, poorly formatted and ​unstructured documents.

Automated Updates for 3rd Party Packages: Because ​portions of the documentation referenced 3rd party tools ​and software used by the company, we wrote a script that ​would compare the version number of the software ​referenced internally with the latest version and flag articles ​for review so that information would not remain too out of ​date.

The Results

Not only did we save this company a ton of time and money ​using a combination of unsupervised learning and classic ​automation techniques, we also coached them on how to ​institute a guild model for shared governance across their ​teams to ensure everyone is invested in the long term ​maintenence of their internal documentation.

Case Study #2: Better Fraud Detection & Consumer Insights

The Challenge

A rapidly growing start-up brand specializing in streetwear ​drops reached out to us because they wanted to better ​detect bot fraud and optimize their drop timings. They had ​accumulated a vast amount of unstructured data and ​weren’t sure what to do with it.

Our Approach

Data Lake Assessment & Reachitecture: We assessed all ​sources and the quality of data in their data lake to refine ​their data pipeline going forward. This included renaming, ​recategorizing, and normalizing the datatypes in the data ​lake, as well as assessing their (lack of) indexing.

Collaborative Filtering for Better Consumer Behavior ​Insights: Collaborative filtering is a recommendations ​technique that predicts user interest by assuming that ​people who agreed in the past will agree in the future. We ​used this method to analyze behavior patterns, preferences, ​and interactions so we could provide personalized upsells ​and marketing strategies on a per-user basis.

Time-Series LSTM Neural Net for Forecasting Trends: An ​LSTM (Long Short-Term Memory) is a type of recurrent ​neural network (RNN) designed to recognize patterns in ​sequences of data, with the ability to remember long-term ​dependencies. LSTMs are great for forecasting consumer ​trends and identifying potential high-demand items for a ​retail company

Advanced Fraud Detection Using Isolation Forests: An ​Isolation Forest is an anomaly detection technique that ​identifies outliers by isolating them through random ​partitioning, using fewer splits for anomalies than for normal ​points. This helped us identify bot behavior more easily.

The Results

By re-architecting this company’s data lake and ​implementing machine learning solutions that enabled this ​rapidly growing brand to gain valuable consumer insights ​and improve fraud detection, we helped them boost their ​sales enough to reach several internal milestones in the ​quarter following our engagement. We think that’s pretty ​sweet.

Case Study #3: Getting Big Results From Small Datasets

The Challenge

A music start-up that had recently raised it’s Series A came ​to us and asked what their options were for creating a better ​recommendation system for their digital products. They ​didn’t believe they had enough data to do this, but because ​they were dealing in music assets, we thought differen​tly.

Our Approach

Repurpose Open Source Datasets: We created a ​composite dataset of some of the songs from the GTZAN ​and FMA music datasets made up of music that we knew ​matched the genres of their dataset, and partitioned all ​items in the new dataset into shorter samples for ​processing.

Transfer Learning: Because we only used a fraction of the ​relevant songs from the open source datasets, we then ​trained several neural network models using these sets, and ​then re-trained the last few layers using our revised, ​repuposed dataset that included their original music to train ​the models.

Keras Optimizer: We used the Keras Optimizer package to ​tweak and refine the models we built, including optimizing ​the hyperparameters. We didn’t leave anything to ​guesswork, which we’ve unfortunately come across a lot in ​the projects we have inherited from other developers.

Euclidean Distance to Find Similar Sounds: We used the ​neural network to disciminate between genres, and then for ​the genre the music was most likely to be in, we found ​similar songs in that genre using a classic machine learning ​technique using euclidean distance.

The Results

This company told us that we had gotten them further than ​any of the folks they had hired before for this task, which ​was music to our ears. The company had plans to use the ​tool manually for curating new songs, and also to put it in ​their product in a mobile app that lands musicians sync ​placements and industry connections.

A lot of machine learning shops out there use ​off-the-shelf solutions with minimal ​understanding of the internals.

Not us. You want the latest and greatest? We’ll ​do the research and implement from the latest ​papers for you.

What Makes Us Different

Artificial Intelligence

Keras, Ternsorflow, & more

Whether Tensorflow and Keras, PyTorch, ​Scikit-Learn, or somethign else, we can ​stitch together the right stack of tools to ​get you great results.

We want to spend our time solving your ​problem, not sitting in meetings.

Deep Learning Line Icon

We Don’t Like Guessing

We use tools like the Keras Optimizer, ​stratified sampling, and K-fold cross ​validation to make sure we are getting you ​the best results possible.

Machine Learning doesn’t have to feel like ​a mysterious black box.

Cycle icon

ML/AI Explained Simply

In our milestone presentations, we break ​down the machine learning lifecycle so you ​understand what we are doing and what to ​expect.

We train your team so they feel confident ​in how to maintain your new AI investment.

Transparent Liquid Blob

Our ​Process

Intro Call &


We run through our systematic checklist so ​that we get you exactly what you need for ​your specific business case.

We Build Your ​Solution

We use industry best practices to deliver the ​solution as specified, tailored to your needs, ​following the machine learning lifecycle.

We Train & Transfer

We teach you and your team how to take care ​of your new machine learning model and how ​to put it into production effectively.

Code Logo

En​gram Software

Let’s chat

It’s like brainstorming with friends – but these pals ​can help your business take advantage of machine ​learning and AI to stay ahead of the curve.