Resume

Shahin Taghikhani

Applied Scientist III, Amazon DSP

Applied Machine Learning Scientist with a track record of taking products from 0→1 fast by turning vague business problems into concrete math, code, and shipped systems. I specialize in large-scale adtech, combining control systems, machine learning, and system identification, with hands-on experience building truncated Fourier series models for seasonality, multi-agent troubleshooting frameworks, real-time throttling, and deal allocation models. Persistent and execution-focused, I iterate quickly with stakeholders and harden successful ideas into reliable, high-throughput, revenue-driving production services.

Career

Experience

Amazon Demand Side Platform (ADSP)

Applied Scientist III

New York, NY

Sep 2022 – Present

  • Designed RADAR, a multi-agent troubleshooting framework trained on historical tickets to diagnose cross-stack DSP issues, reducing resolution time by 20% and improving reliability during peak events.
  • Built a seasonality compensator using truncated Fourier series to optimize bidding and allocation with high interpretability, shrinking model size 6x and cutting CPA by 4-6%.
  • Delivered a Model Control Monitoring Tool with minute-level observability across 15-20M TPS RTB traffic, driving $4M Prime Day revenue uplift.
  • Created Swift Throttler bid blocker with cached inference under 10 ms, cutting wasted bids, trimming infra cost by roughly $1M, and improving win rates by about 5%.
  • Developed a PMP vs OA deals model (MLP + PID control) with cached gamma CDF inference (~15 ms) shipped through SageMaker pipelines and Lambda for rapid iteration.

Aetna / CVS Health

Senior Data Scientist

New York, NY

Sep 2019 – Sep 2022

  • Led DashForce experiment measurement and monitoring framework (5 people) using bootstrapping/CUPED and causal inference; estimated $5M annual savings.
  • Built a multi-armed bandit subject line optimizer that increased open rates by 11 percentage points.
  • Shipped WoundDetection computer vision web app MVP with estimated $7M medical cost savings.
  • Led depression detection campaign using deep learning and NLP on call transcripts and claims data, saving $4M over a 6-month pilot.
  • Drove readmission avoidance analytics reducing 600+ readmissions (about $8M in savings).
  • Created LocalExpert location recommender increasing targeted members 10%, behavior change 2%, and ~$6.4M in revenue.

Data Intelligence Lab, East Carolina University

Research Assistant

Greenville, NC

Aug 2017 – Sep 2019

  • Built YourPersonality ensemble web application on social media data, outperforming prior results by 2.7%.
  • Conducted applied research across NLP and computer vision with a focus on scalable, privacy-aware modeling.

BulletProof.com

Data Analyst

Seattle, WA

May 2018 – Aug 2018

  • Developed an end-to-end material requirement planning system that reduced out-of-stock issues, saved about $100K, and lifted attributed sales by 2.8%.
  • Built real-time warehouse and transportation KPI dashboards improving efficiency by 35%.
  • Led inventory discrepancy root-cause analysis, improving accuracy by 15% in two months.

THINKCO.ir / Shopit

Founder / Product Supervisor

Tehran, Iran

Aug 2016 – Aug 2017

  • Ran SEM/SEO programs for 25+ small businesses, improving conversion rates by roughly 35%.
  • Implemented market segmentation and A/B testing that cut CPC by 30% and boosted acquisition by 15%.
  • Raised $50K seed funding for Shopit, optimized UI via statistical analysis, and built marketing analytics products.

Takhfifan.com

Sales and Marketing Analytics Supervisor

Tehran, Iran

Jan 2016 – Aug 2016

  • Deployed an unsupervised customer segmentation framework, growing B2B accounts by 30% and adding $100K quarterly revenue.
  • Ran marketing experiments across email, UI, and creative to increase conversion and open rates.
  • Led competitive analysis and omnichannel campaign planning to improve pricing, positioning, and placement.

Foundations

Education

The Data Incubator, NYC

Data Scientist Fellowship

Apr 2019 – May 2019

  • Awarded fellowship

East Carolina University, NC

Master of Science, Software Engineering

Aug 2017 – May 2019

  • GPA 4.0
  • Awarded full scholarship

University of Tehran, Iran

Bachelor of Science, Industrial Engineering / Operations Research

Aug 2008 – May 2013

Builds

Projects

Deal Score Ranking Overhaul

Replaced PMP waterfall ranking with win-rate-based Deal Score models to maximize advertiser surplus across PD, PA, and OA bids while enabling multi-bidding and smarter bid shading.

Win Rate ModelingAuction DesignOptimization

RADAR Troubleshooting Framework

Multi-agent system coordinating specialized models to diagnose ad stack tickets and recommend mitigations.

Machine LearningMulti-agent SystemsObservability

WoundDetection Web Application

Visit

Computer vision MVP to detect skin conditions and track member progress with estimated $7M annual savings.

Computer VisionDeep LearningPython

YourPersonality Web Application

Visit

Ensemble personality prediction using social media signals; beat prior benchmarks by 2.7%.

NLPEnsemble ModelsFlask

LocalExpert Location Recommender

Unsupervised learning engine recommending optimal sites of care, improving targeting and conversion.

Unsupervised LearningRecommender Systems