AlphaCX NLP System
NLP Engineer
About the Project
AlphaCX is a customer support platform where I built the NLP system responsible for automatically classifying incoming support tickets and analyzing customer sentiment. The system routes tickets to the right team based on content analysis and flags urgent negative-sentiment cases for priority handling.
What It's Made Of
Classification pipeline
Text preprocessing · domain-trained classifier
Takes an incoming ticket and categorises it so it can be routed to the right team. The preprocessing is the substantive part: support language arrives as abbreviations, typos, mixed languages and emotional writing, so normalisation was built against the real ticket corpus rather than borrowed from a general-purpose NLP pipeline.
Sentiment analysis
Frustration & urgency detection
Scores the emotional register of a ticket to detect customer frustration and urgency, which is a different signal from what the ticket is about. A polite billing question and a furious billing question route to the same team but not with the same priority.
Priority scoring & routing
Combined signal · smart routing
Combines category and sentiment into a single priority score that drives routing, so urgent negative-sentiment cases surface for immediate handling instead of queueing behind older but calmer tickets. This is where the measured improvement in response time came from.
Backend integration
Node.js · AdonisJS
Wired into the existing AdonisJS support platform so classification happens on ticket ingestion with no separate step for an agent to remember. The NLP system is a service the platform calls, not a dashboard someone has to visit.
Key Highlights
- Built text classification pipeline for automatic ticket categorization
- Implemented sentiment analysis to detect customer frustration and urgency
- Designed priority scoring system that combines classification and sentiment for smart routing
- Integrated with the existing AdonisJS backend for seamless ticket processing
- Reduced average ticket response time by improving routing accuracy
Technical Challenges
Customer support language is highly informal — abbreviations, typos, mixed languages, and emotional writing. Building a classifier that handled this noisy input reliably required extensive text preprocessing and training on domain-specific data rather than relying on general-purpose NLP models.