Shane Thakkar

A live analytics platform that grades every NFL player on a 0–100 scale across 12 positions, using statistical models tuned per position. 10 seasons of play-by-play and tracking data flow through a Python pipeline into Postgres; the web app surfaces per-season grades, career trends with uncertainty, and full depth charts for all 32 teams.

Pythonnext.jsPostgreSQLvercelnflverse

visit the site

Flagship — Live

prompt-data.vercel.app

Confidently-Wrong Rate●

Baseline (no trust)100%

Prompt Data41%

Over-decline (clear)0%

ECE 0.44 → 0.15EXEC ACC 55.8%

Text-to-SQL · Trust Layer

Prompt Data

Ask an e-commerce database a question in plain English, and trust the answer. The product is trust, not query generation: it asks instead of guessing on ambiguous questions, surfaces the assumptions behind every query, and attaches a calibrated confidence score. A reproducible eval harness on the BIRD benchmark backs every claim, cutting confidently-wrong answers from 100% to 41% while over-declining none of the clear questions.

PythonFastAPIClaude APInext.jsSQLite

visit the site

Flagship — Live

venue-insights.shanethakkar.com

Category Performance

●

F&B Analytics · Fully In-Browser

Venue Insights

Every venue exports its point-of-sale data differently and messily; this turns any of it into the same clean, documented, operator-grade analytics. It infers the schema from arbitrary headers, runs a transparent cleaning pipeline with a 0–100 Data Health Score, then produces ~13 analysis modules — category performance, Pareto tail-SKUs, menu engineering, market-basket lift, staffing curves, price elasticity — plus a plain-English executive summary and an exportable methodology report.

TypeScriptnext.jsWeb WorkersPapa ParseSVG viz

visit the site

/ 02

Projects

05 PROJECTS

Published · JUN 12 '26Privacy · Regulated Data · DuckDB

I Built the Missing Piece of the Ride-Hail Data Fight

Regulators want trip data; riders don't want to be tracked. RideCloak is the pipeline in the middle — it finds the personal data, generalizes locations until 90% unique trips fall to 0.06%, and logs every step in a tamper-evident ledger. Plus an AI agent that can't be talked into leaking.

pythonduckdbnlpk-anonymityllm-agents

read article

Published · MAY 25 '26SEC · NLP · 10-K Analysis

Failing Companies Tell on Themselves in Their Annual Reports

A model that reads what companies say about themselves in SEC filings catches 79% of bankruptcies across 24 cases. The 'false positives' include Nordstrom, Walgreens, Macy's, Kohl's, CVS, and Lucid Motors, all of whose decline didn't end in court.

pythonsklearnpandasnlpedgar

read article

Published · APR 22 '26NFL · WPA · XGBoost

Fourth Down Is Still Football's Biggest Coaching Problem

107k decisions from 1999–2025, scored against historically optimal calls. Coaches still leave ~one free win on the table every year — and the conservative ones make most of their mistakes in the red zone.

pythonpandasnumpyxgboostnflfastR

read article

/ 03

Tech I work with

STACK

drag to pick up

python

sql

jupyter

pandas

scikit-learn

tensorflow

keras

snowflake

databricks

docker

tableau

power bi

excel

react

next.js

claude code

/ 04

About

WHO

Hi, I’m Shane.

AI and automation builder working with Highgate Hotels’ acquisitions team, developing tools that support deal evaluation and reduce manual work across the deal pipeline.

Full stack across Python, SQL, TypeScript, and Next.js, from data pipelines through production deployment. Outside of Highgate, I build independent projects in sports analytics, prediction markets, and applied machine learning.

Quick Facts

[ 05 ]

Education: B.S. Business Analytics & AI, UT Dallas (May 2026)
Based: Frisco, Texas
Focus: Analytics · Data Science · ML · BI
Stack: Python · SQL · R
Teams

/ 05

Get in touch

LET'S TALK

Always up for the next problem worth solving.

Happy to connect about projects, ideas, or interesting problems in AI, ML, and analytics.

get in touch resume.pdf

Emailshane.thakkar@gmail.com

LinkedInin/shanethakkar

GitHub@shanethakkar