Barath's Portfolio

Hi! I am a Forward Deployed Engineer (GenAI) at Google in New York City. I recently completed my Master of Business Analytics at MIT, where I was a Graduate Research Assistant advised by Professor Vivek F. Farias. Before that, I received my B.Math in Statistics & CS from the University of Waterloo, where I was mentored by Professor Steve Drekic. Outside of work, I'm usually building something: agents, eval tools, small experiments with language models. I'm also a huge basketball fan! Please reach out!

Research

My research so far spans generative modeling of digital twins, behavioral simulation using large language models, and token entanglement in language models. I am interested in how large language models reason about human behaviour, and what their hidden representations mean for interpretability and safety! My recent work includes synthetic customer modeling at Anheuser-Busch InBev through MIT, and independent work on how models pass along preferences through data that seem meaningless.

Work Experience

Forward Deployed Engineer, Generative AI · Google | Sept 2026 – Present

Deploying Google's frontier models in customer environments.

Machine Learning Engineer · Wayward (Series B) | Feb 2026 – Aug 2026

Brand–partner matchmaking and monetization modeling for Wayward’s partnership platform.

Shipped the live brand-publisher discovery-scoring model (LightGBM, 1,123 pairs scored) over 14.2M ad-spend rows ($247M GMV) across 8 publishers and 590 brands; engineered 2 external Google search-volume signals to score cold-start brands with no campaign history; AUROC 0.68 (0.73 with prior history), validated per-publisher on a temporal holdout
Owned the ROAS-based labeling and feature pipeline; caught a $16M attribution blind spot (48 brands) from a case-insensitive join bug before it corrupted training; rebuilt the model's top feature to lift coverage from 2.5% to 98.8%

Graduate Research Assistant · Massachusetts Institute of Technology | Sept 2025 – Aug 2026

LLM-based digital twins and behavioral simulation for ABI’s Zé Delivery (5M users) platform.

Developed an AB InBev digital-twin pipeline converting 319 Brazilian-Portuguese interviews into 291 queryable personas
Built a GEPA prompt-evolution loop to learn 20 behavioral rules per persona, replacing manual prompts with persona cards
Deployed an LLM-as-judge evaluation harness over 5,590 held-out turns with a 4-dimension rubric, scoring 4.72/5.0 overall

GenAI Lab Team Member · BMW Group (Innovation & Research) | Feb 2026 – May 2026

Implementing a feedback-driven prompt optimization pipeline for continuous AI agent adaptation without retraining.

Created a prompt optimization pipeline for OCR extraction, improving accuracy by 54% (0.437 to 0.673) over 5 iterations
Built an evaluation layer, fixed scoring bugs, and used model-vs-prompt ablations (+0.188 vs +0.025) to guide refinement

Analytics Lab Team Member (2025 Winners) · Mira Intel | Sept 2025 – Dec 2025

Computer vision & multi-agent workflow for wind-turbine damage detection and forecasting.

Awarded 1st Place in MIT Analytics Lab (top team among 22 teams across 21 industry partners) [official announcement]
Constructed a computer vision pipeline that ingests images of wind-turbine blades, predicts damage, and forecasts wear
Engineered a YOLOv8 damage detector using Python achieving 0.96 F1 on critical defects via class-imbalance correction
Architected a multi-agent workflow using OpenAI Agent Builder to convert lengthy damage reports into prognostic JSONs

Software Engineer Intern · PointClickCare | Sept 2024 – Dec 2024

A full-stack platform for the Senior Living Resident Management market (React & Python).

Built a feature enabling an Azure GPT3.5 Turbo model to review lengthy SQL files and flag syntax deviations in pull requests, providing real-time non-intrusive quality checks using Groovy (i.e., undisturbed Jenkins build)
Took ownership of complex React UI features such as debugging production issues, building a multi-select search filter, and enabling editable time fields for “move-out” operations

Software Engineer Intern · Manulife | Jan 2024 – Apr 2024

Qualys Cloud Agent automation in Ansible using Azure Metadata and YAML.

Proved out Qualys Cloud Agent functionality in Ansible by leveraging Ansible Modules, Azure Metadata, and YAML
Built a Chef feature using Ruby to validate access control policies (Sudo, HBAC) during the provisioning of a server, strengthening security and reducing server deployment time by 15%

Software Engineer Intern · Manulife | May 2023 – Aug 2023

PySpark and Databricks pipelines for large-scale database migration (80K records).

Transformed data fields using PySpark and Databricks to support large-scale database migration of 80,000 records
Refactored a mission critical React project by consolidating logic, modularizing code, and deleting files
Built API-driven table rendering functionality (< 3s response) using TypeScript, React, and Material UI
Streamlined client data entry by adding button functionality and 5 conditional fields using React Hooks and JSX

Software Engineer Intern · Manulife | Sept 2022 – Dec 2022

Azure Logic Apps integration of Manulife Bank (SOAP) and Salesforce (REST).

Integrated Manulife Bank (SOAP) with Salesforce (REST) using Azure Logic Apps, achieving 42% more performance (less than 1s latency) and near-zero downtime compared to the previously deployed PCF model
Solved authentication using Azure App Service to allow the Logic App to hit the bank’s on-prem SOAP APIs
Strengthened Logic App security by implementing OAuth 2.0 authentication using Azure Active Directory

Software Engineer Intern · PointClickCare | Jan 2022 – Apr 2022

Azure Event Hub deployment & Python multithreading for large-scale (22M+) data ingestion.

Provisioned an Azure Event Hub using Java and Terraform to enable parallel ingestion of 22 million patient data
Leveraged Python multithreading to reduce script validation time for OHDSI datasets with 150,000 rows by 56%
Programmed an Azure Function using Java to detect the health of a Redis cache and PostgreSQL application

Software Engineer Intern · PointClickCare | May 2021 – Aug 2021

API development and unit testing using Spring and JUnit.

Engineered a new API in 1 week (half a sprint) using Spring to inform 10,000+ vendors on ancillary charge statuses
Integrated 20+ comprehensive unit tests with JUnit for public APIs to improve API robustness and maintainability

Selected Research Reports

Projects

Extracurricular

Clubs

MIT AI & ML Club, MIT Tech Club, MIT Project Management Club, UW Data Science Club, Stanford ASES

Conferences & Events

CODE@MIT (2025), GAI World (2025), Hack The North (2024), Jane Street FTTP (2021)

Interests

strategy games (e.g., Chess), lo-fi music, long road-trips, and electronic music production!