TCS Research · Feb 2019 – Mar 2022
NLP query chatbot
Natural-language issue search and next-word prediction, at research-lab scale.
Applied Machine Learning Engineer
3.2 yrsapplied ML research
Sparkdistributed training & prediction
The challenge
Users needed to find known issues by describing them in their own words, and models had to train and predict at a scale a single machine couldn’t handle.
How it works.
NLP issue search
A query chatbot built with Django uses NLP to match user questions to relevant issues.
Next-word prediction
An LSTM sequential model (Keras, TensorFlow/PyTorch) predicts the next word as users type.
Distributed ML and MLOps
Large-scale training and prediction ran on Hadoop and Spark, with A/B testing and MLOps managing the model lifecycle.
The outcome
A working NLP chatbot and prediction module, trained and served at scale, with experiments and releases managed through A/B testing and MLOps.
My role
- Built the chatbot, the LSTM module and the distributed training pipeline
Stack
- Python
- Django
- Keras
- TensorFlow
- PyTorch
- LSTM
- Hadoop
- Spark
- MLOps