Ali H. Naeini, PhD

Experience

• 

  Digit

  May 2019 - Jul 2019

  Data Scientist

  Purchasing Time and Value Prediction using Time Series (- Goal: Forecast the next purchased date and monetary value of 375K customers-Dataset : 1 million rows and 3 Columns of CustomerID, InvoiceDate and Unitprice.- Outcome 1: The frequency of the customers transactions for each customer evaluated separately andtransaction dates are forcasted.The accuracy of the xgboost model is 75%.- Outcome 2: The transactions’ value are predicted using unsupervised learning(K-means) and LSTM with 80% accuracy Show less

• 

  Insight Data Science

  Sept 2019 - Nov 2019

  Artificial Intelligence Fellow

  Training Environment for Autonomous Vehicle using Reinforcement Learning–Usedpixel and environment parametersas the input to the Deep Reinforcement Learning model–Environment, Observations, Actions, Reward functions are defined in Unity (written in C#).–A backend bridge of C#-Python defined using Unity ML agents to use the state-of-art model-free on-policy Deep Reinforcement Learning model for continuous action space(Proximal Policy Optimization(PPO))–Multi-agent training is completed using AWS EC2-P2 xlarge instance [training speed = 100K episode/hour Show less

• 

  Pathmind

  Nov 2019 - May 2020

  Reinforcement Learning Scientist

  Pathmind is a SasS company which provide service to business for optimization of Industrial Processes using Reinforcement Learning:– Prototyped a continuous learning engine to learn from real experience and plan with a simulation.– Successfully Applied the Proximal Policy Optimization(PPO) Algorithm to reduce the electricity cost ofan Australian factory by 18% compared to the baseline(random action).– Improved the backend bridge between Anylogic(Written in Java) to RLlib(written in Python). Show less

• 

  Einstein Grid

  Jun 2020 - Oct 2021

  Co-Founder and Machine learning Engineer

  – AI Model Research and Development: I designed a peer-to-peer EV Sharing platform where owners share and EVs are delivered by the shuttle drivers to renters. All the shuttle operation are automated by the combination of Deep Learning and Reinforcement Learning models.– AI-embedded App Development: I Developed two IOS apps where AI model(written in PyTorch framework) was directly embedded in the apps in order to reduce latency and preserve privacy.

• 

  Insider, Inc.

  Oct 2021 - May 2023

  Senior Data Scientist

  • Using Transformers include ChatGPT for Topic Authority– Goal: Understand google rank of insider articles against competition in various Topics.– Dataset: More than 1 TB data from Insider Articles’ metadata plus their current google rank.– Action : Tested various Transformers Models include ChatGPT(GPT-3.5) and BERT to create embeddingof each article metadata string and find similarity versus other articles which caused clustering of articlesinto Topics(Cars, Fruits, Robots etc) automatically. Afterward, assigned a Topic Authority Score to eachcluster(category) using their google search rank multiplied by click volume of all articles in that category.– Outcome : After approved by Editorial team, these model put into production pipeline. It significantlyincreased the Overall Traffic/Ad Revenue by focusing on articles with highest Topic Authority in 2023.• Supervised learning for Article Update Prioritization– Goal: Automating the process of prioritizing most important articles to update.– Dataset: Wrote SQL for Querying multiple larges tables(100+ GB) to create feature tables in Snowflake.– Action : Using more than 60 features and Gradient Boosting model, Update Score created for all thearticles two month in advance, so the editorial team have enough time to update the article that gets mostattentions from users.– Outcome : Since Editorial was able to focus on the update of articles with most important articles tousers, Business Insider traffic and its ad-revenue increased about 10% compared to similar period last year. Show less

• 

  Spotter

  Jul 2023 - now

  Senior Applied Scientist (LLMs, Diffusion, RLHF, Alignment)