Barinder Thind

i work as a staff data scientist @ . my interests:

nlp
functional analysis
signal processing
walking
tennis
chess
overwatch
watching movies
listening to music
yugioh
carnivorous plants

i have a book coming called In ML: preorder ↵

feel free to scroll around :)

cv

my resume

download ↵

open full pdf ↵

work

staff data scientist @

work —

where i currently work. it’s varied a lot — a summary:

pioneer member of the data team.
stood up the data stack from scratch — warehousing, reporting db, etl, productionizing.
grew the team: hiring, mentoring, okr-setting, and tooling.
cross-functional work with customer experience, dev, and executive teams.
company-wide financial & customer forecasting (arima, prophet).
nlp in production — brand/creator matching, similarity scores, topic modelling (word2vec, doc2vec, glove, bert, gpt, openai api).
private python libraries + docker containers for model deployment.
deployment via flask & sagemaker; automated by circleci, helm, argocd.
rshiny & flask apps for internal ml servicing.
custom reforge methods to isolate user behaviour (setup, aha, habit moments).
primary owner of a/b test validation & experimental design.

modelling

nlp (word2vec, lda) supervised (regression, trees, neural nets) unsupervised (pca, clustering) monte carlo

programming

r python sql ruby js / react

software

aws (sagemaker, s3, ecr, ec2, lambda) amplitude bigquery gce fivetran segment rshiny flask ga notion zendesk asana lm studio ollama hugging face + more

at the moment, i’m building some new product features.

archive

some prior projects

astromlfinancial ml web app ↵

superscribefast speech recognition · soon

gaussian processesa writeup ↵neural odesa writeup ↵least-angle regressiona writeup ↵

nlpbegin

a course through the math of nlp, openly available (circa 2024)

learn nlp

large language models

current standard

frontier

open nlpbegin ↵

In ML

a machine learning book where i ramble a lot, made with love!

[ summary + chapters — placeholder ]

preorder (soon) ↵

Deep Learning with Functional Inputs

deep learning with functional inputs

most neural nets expect a vector of numbers — but a lot of real data is better understood as a function: a curve over time, a spectrum, a trajectory. this work teaches feed-forward networks to take functions as input (scalar response, one+ functional covariate, any number of scalar covariates). the nice part: the learned parameters are themselves functions that evolve during training, so the model stays interpretable. it holds up on prediction and on recovering the true coefficient function — confirmed via cross-validation and simulation.

read it at amstat ↵

funcnn — an r package

to the best of our knowledge, the first package in any language for deep learning on functional + scalar covariates. built on keras / tensorflow, it gives you simple functions to build models, predict, and cross-validate — plus a write-up of the underlying methodology.

package site ↵