I build teams and deploy ML / deep learning / RL products at scale
conversational ai at scale
As Co-founder and CTO of NextGenVest I conceptualized, built and deployed an AI system capable of extracting data (LSTMs) from open-domain SMS conversations, engaging users via reinforcement learning and suggesting chat responses in real-time chats (deep learning).
Conceptualized and planned full product development.
Built first version to handle over 3million messages for over 60k students.
Hired and managed team of 5 engineers and a product manager to continue platform development.
Guided key technical decisions (cloud infrastructure design, front-end choices, back-end API design).
Designed and implemented data processing and model deployment pipeline.
The engineers I hired are now senior org leaders at the NGV Parent company (VP of eng, senior engs).
As ai researcher
Identified best ML approach for our dataset scale and use case.
Implemented and neural network based on Google Smart Reply (Tensorflow open-sourced).
Set up data processing pipeline to generate responses in < 200ms.
data extraction and user engagement at scale
i conceptualized a system to drive costumer engagement using reinforcement learning
For any given user, the RL system looks at conversions and long-term engagement for similar users. The model estimates the expected reward of a particular user given all possible messages we could send them. Finally, it sends the message with the highest expected reward to the user. The video above explains this in more detail.
data extraction at scale
Another big challenge was extracting relevant information form text conversations to help users fill out forms, and access financial aid resources. I approached this solution by fussing recurrent neural networks with a UI optimized for humans to provide training data during their normal course of operations. See the video above to see how this works.