Customer Sentiment Classification
in Realtime


Project Overview

Ensuring customer satisfaction throughout their journey is crucial in preventing churn, yet the level of personalized response required to maintain a positive customer experience is not scalable in high volume e-commerce environments. Automated screening of customer sales-person interactions is possible with Natural Language Processing and Deep Learning. Quantum Analytica specializes in creating custom language models. This is why Avochato chose Quantum Analytica as a partner to build a real-time sentiment classifier for their communications platform. We worked with them to create custom language embeddings geared towards SMS, Slack, and colloquial/informal semantic structure in order to classify how positive or negative customer communications are in real-time.

Our Deliverables

Custom Language Model

Sentiment Classifier

Scalable API Endpoints

Project Goals

Avochato is a successful Bay Area company that is a hub for all modalities of business communication. Because of the high frequency use of their platform, developing a low-latency API endpoint for sentiment classification was essential. We worked with Avochato from the ground up, where we critically examined the linguistic and semantic structure of their data. We built a custom language model, using the latest advances in transformers, to drive a self-supervised model that robustly handles a widge range of sentiment including: level of positivity, level of negativity, emotion handling, and sarcasm detection. In order to prevent latency, we deployed the solution as a Django API endpoint in a docker container that was hosted on a cloud platform for auto-scaling.

Our Approach

When developing Natural Language Processing models, it is essential to understand the underying linguistic structure of the data. We focused on ensure contextual understanding of the language is preserved in the modelling process, which is why we chose to implement a transformer based solution. Our pipeline focused on using transformers, coupled with a self-supervised deep learning network, in order to classify sentiment.


Utilizing slack, email, and SMS communications, we built a realtime sentiment classifier to monitor customer service exchanges in order to quickly identify customers who had a negative experience, so that a customer engagment specialst could quickly identify and improve their customer journey.

Client Feedback


“Our team at Avochato was blown away by Quantum Analytica’s depth of knowledge in AI/Machine Learning and creativity in helping us model our data and find the answers we were looking for. It only took a few conversations to guide our MVP and I stepped back and let then work their magic. Quantum Analytica was prompt and courteous with project updates, and was quick to respond while we iterated on the data model. This helped our team stay agile and produce a functioning and deployable classifier on a tight deadline!

Christopher Neale

CTO, Avochato