Customer: The Client is a Bay Area based technology company that transforms the way
enterprises and companies deliver training to their workforce.
Problem: Non-technical founder of this startup wanted us to develop a HR productivity
solution that reflects what the next generation workforce should look like, that is, more
connected, mobile, and social.
Solution: ValueLinked researched, prototyped, and developed a R1.0 of the product using
Ruby on Rails, AWS and MangoDB using a team of 5 people (Onsite and Offshore based).

Customer: E-Commerce Company based out of Austin TX. This $100M company ships
memory and storage upgrades and accessories for the Apple Macintosh to nearly 200
countries.
Problem: Client’s call center fielded inquiries and issues from customers using calls and
chat. Client had text files containing several years of chat logs. Client wanted a bot to
offload simple tasks from their call center to free agents for more difficult more
technical interactions.
Tasks: 1) determine feasibility and scope of project and 2) implement a bot.
Actions: Did data exploration and analysis of text. Identified a collection of the most
common topics, questions and problems. Talked with agents and scoped the project to
cover initial low hanging fruit – easy tasks the bot could handle to take some load off
agents.
Results: developed prototype bot integrated with client’s product database and scaled it
for production.

Customer: One of the Leading chip manufacturers in the world who needed to eliminate
defects in wafers using images of the wafers.
Actions: Used YOLOX and deep learning on images for object recognition and
registration. Deployed models into deliverables: put in Docker, fired up 3 Docker
instances, each accessed by a separate API:
1) ML Pytorch using Flask to provide API for this container
2) postgress with API
3) Django server providing UI for the whole system, making API calls to DB.
Only server 3 used DB calls. The ML instance went thru Django to get to the db. This
supported DB schema consistency. APIs were used to start training, called when
finished, and supported estimated time to finish. Images were stored in shared files.

Client: One of largest postal services organization in the world if not the largest.
Situation: The client was running mail management services, 20 billion operations per
day in Cassandra database.
Actions: Developed never stop pipeline for post office, even during upgrade. Developed
system only rolled forward. System had no delays during writing and reading. All
frequent operations were short. conditioned data such as zip code normalization.
Results: 20 billion operations per day reliably, non-stop, brought data though a large
100+ node Cassandra database. Presented at Cassandra Summit.

Customer: The Client is a US$ 19.2B American Multinational Personal Care corporation
who produces sanitary paper products, surgical and medical instruments.
Problem:
• Client’s data lake contained hundreds of databases, not well structured. The Client
wanted to move away from SQL Server in Azure to Snowflake on Azure.
• Client was on Azure using React, Node.JS, Databricks and Snowflake. Client wanted
every day deployment which was not supported by current architecture.
• Client was on Azure using React, Node.JS, Databricks and Snowflake. Pipeline was:
output to datalake -> excel -> snowflake -> app API. pipeline was not so good:
manually triggered and not fully automated.
• The CD frequently broke, the client did not have Linux, Java or Tomcat skills to
diagnose or fix these problems quickly.
Solution:
• We created a fully automated version of the pipeline was created. Java and Spring
Boot were used to create APIs.
• The model ran in Databricks and output went into Databricks and then moved into
Snowflake.
• Started with dirty, disorganized data, incompatible schemas. CI/CD was originally on
Atlassian but was moved to Azure and based on Git using Java and Tomcat with a
webapp inside Docker.

Customer: One of the largest mobile communications companies in the world with
Android Mobiles manufacturing and distribution business
Problems our engineers worked on:
■ Created mobile applications focused around machine learning and natural language
processing.
■ Started and led development of a recommender system and pushed it to the Client’s
Android phones
■ Led natural language processing efforts focused on applications like prioritizing
incoming messages
■ Prototype software inventions, built Find-Wipe-Lock app for the Client’s mobile
phones using AWS SimpleDB for service all continents.
■ Term and Image database for global development of games for the Client.

Customer: This $8B Medical Systems Company out of Palo Alto CA. They make linear
accelerators and software for treating cancer and other medical conditions with
radiotherapy, radiosurgery, proton therapy, and brachytherapy.
Problem: Initially asked by Director of Engineering to help evaluate third party providers
of machine learning solutions, then asked to help develop predictive maintenance for
thousands of machines distributed globally.
Solution:
• We wrote Python code to extract data from logs.
• Conducted exploratory analyses using R.
• Did feature engineering and wrote R code to generate derived features to facilitate
anomaly detection.
• Used machine learning to generate comprehensible models that predicted failures
ahead of time.
• Wrote Python code to distribute processing over virtual machines in Azure.

Customer: One of the largest media and entertainment companies in the world with
US$ 12 Billion in revenue.
Problem: The Client has a user base close to 100 million users worldwide. Client wanted
us to build an analytics platform that tracks user’s behavior whenever an upselling
opportunity is presented to them.
Solution: Our consultants were responsible built the front-end of a web application and
on completion of the front end, we completed the integration with the backend. We
used a team of onsite and offshore team members.
Technologies involved: Python, JavaScript, PostgreSQL