cursor
stg
ES Contact Contact

ContactLoop

Data engineering and AI for conversational intelligence

Service:

Data engineering and machine learning

Client:

ContactLoop

Industry:

SaaS · conversational intelligence

Year:

2024

Duration:

10 months

Data and artificial intelligence architecture for an Australian conversational intelligence startup. A system that predicts customer intent before they finish speaking, changing how companies handle millions of interactions.

ContactLoop — conversational AI platform

The challenge

ContactLoop processed conversations with generic models that didn't understand each industry's context. Intent prediction accuracy was 52%, barely better than chance.

They needed a data pipeline able to train vertical-specific models (banking, insurance, telco) with their clients' proprietary data.

  • Model training pipeline per business vertical
  • Real-time processing of 100K+ conversations a day
  • ML feature system with multiple time windows
  • A/B experimentation infrastructure for models in production

The solution

We designed a data lake with a medallion architecture (bronze/silver/gold) and a feature store that feeds per-vertical models. Each model is trained on industry-specific data from the client, reaching significantly higher accuracy than generic models.

  • Medallion-architecture data lake on Databricks
  • Real-time feature store with Apache Kafka + Redis
  • MLOps pipeline with automatic weekly retraining
  • Drift and performance monitoring dashboard per model

Stack

Databricks · medallion architecture · Apache Kafka · Redis · MLOps with automatic retraining.

Approach

Models trained per vertical with the client's proprietary data, continuous drift monitoring and A/B experimentation in production.

Results

52% → 85%

Prediction accuracy

2.3s → 180ms

Time per conversation

USD 8M

Series A round closed

100K+

Daily conversations

Intent prediction accuracy rose from 52% to 85%. Processing time per conversation dropped from 2.3s to 180ms. ContactLoop closed a USD 8M Series A with these results as a pillar of its pitch.

“STG didn't just build us a data pipeline; they built the competitive advantage that let us raise our Series A.”

CEO — ContactLoop