Global Technology Leader saves 60+% through hyper-automation with LLM based feedback architecture
Objectives
- To transform customer experience using LLM-based hyper-automation solutions
- The current conversational AI solution deployed is trained on limited intents and phrases and is unable to map 50% of the conversations
- A lot of manual effort goes into extracting missed utterances from the bot and mapping them with the right intent to re-train the bot
Customer Environment
- Distributed / unstructured data environment
- Data accuracy and performance is of high importance
- Integrated customer internal enterprise applications
- 1000+ Complex conversational multilingual bots
Solutions
Automate feedback loop with the help of LLMs and existing data science models for classification of missed utterances, information, insights – structured and unstructured . The solution consists of the following key elements:
Download existing intents and training phrases that would serve as model training data.
Embedding Large Language Model – Classification logic based on the embedding data, entity, and utterances; mapping score gets calculated.
ML Model – Created based on LLM data and once trained saved in the client’s cloud for further reference.
Python Notebook – ML and embedding logic will be executed in python notebooks in a scheduled manner.
BOT Training API – Re-training of the bot with new classified data (intent-utterances mapping).
The solution uses LLM power to classify missed utterances to valid intent, contextually.
The entire process is 100% automated with no manual intervention.
Outcome
~90% Automation achieved for conversations
Improve End customer experience enabling contextual conversation with 94% confidence score
Unstructured data processing to build the classification model
Integration with Enterprise Application to automation feedback loop
Agile Methodology for acceleration of development
Proprietary Technology for quick adoption of customer environment to deliver in secured environment
