Chandrahas Reddy Addanki
I design and ship the machine learning and generative AI systems that power decisions at Academy Sports + Outdoors, from where to open the next store to how the company makes sense of millions of product records. 5+ years building production ML systems, currently deepening that work through a Ph.D. in Information Technology (AI specialization).
previously: Illinois Institute of Technology
now: Academy Sports + Outdoors
Background
I started in research: three peer-reviewed publications during my time at SRM Institute, spanning hyperspectral image clustering to IoT-based cardiac monitoring. That work is where I learned to be rigorous about evidence before I learned to move fast.
Grad school at Illinois Tech marked a shift toward applied analytics - dashboards, SQL, stakeholder-facing reporting. Since then at Academy Sports + Outdoors, the work has moved deeper into production ML and generative AI: propensity modeling, real estate analytics, and the internal GenAI tooling - RAG search, LLM data pipelines - the data science team now runs on. A good chunk of it is owned end to end, from the model to the pipeline that keeps it running.
Education
University of the Cumberlands
Ph.D., Information Technology (AI specialization)
In Progress
Illinois Institute of Technology
M.S. Computer Science
Chicago, IL · GPA 3.8
SRM Institute of Science and Technology
B.Tech, Information Technology
Chennai, India
Skills
Machine Learning
Gradient Boosting & Calibration
XGBoost, LightGBM, probability calibration for production scoring systems
Experimentation
A/B testing, uplift modeling, causal impact analysis
Python / R
pandas, scikit-learn, feature engineering pipelines
Generative AI & LLMs
RAG Systems
Hybrid retrieval (BM25 + vector), embedding pipelines, vector databases
LLM Orchestration
LangChain, agentic and multi-agent workflows, prompt engineering
Applied GenAI
NL-to-SQL, LLM-based data enrichment pipelines, production LLM API usage (Vertex AI, Anthropic, OpenAI)
Cloud & Data Engineering
GCP
BigQuery, GCS, Cloud Run, Vertex AI
AWS
SageMaker, S3, Redshift
Pipelines & MLOps
Orchestration, quality-gated retraining, CI/CD for data pipelines
Full-Stack / Product Engineering
TypeScript
Real-time desktop application development
Swift / SwiftUI / SwiftData
Native iOS app development
Firebase
Cloud Functions, App Check, backend-as-a-service integration
Selected Work
Featured Work
01 / Data Science / Retail Analytics
Customer Channel-Migration Propensity Model
The Challenge
The business had no systematic way to identify which customers were likely to shift purchasing channels - meaning retention and marketing spend were being allocated on instinct rather than a defensible, continuously-updated signal.
The Solution
A propensity model - XGBoost with probability calibration, built on ~59 engineered features - scoring the customer base on a recurring weekly cadence. I own the full system, not just the model: a retraining orchestrator on a rolling window, a quality gate that checks each new model against a dynamic, base-rate-relative threshold rather than a fixed cutoff, artifact versioning, and automated failure alerting. Along the way I tracked down and fixed several real production bugs - a boolean-casting error, a regional data distortion in the scoring logic - the kind of issues that only show up once a model is actually running continuously in production.
02 / Data Science / Real Estate Analytics
Store Expansion & Site Selection Analytics
The Challenge
Real estate and merchandising decisions - where to open new stores, how to segment the store network for strategy - lacked a data-driven forecasting and clustering layer to support them.
The Solution
A LightGBM + OLS ensemble for site-level revenue/NPV forecasting, and a K-Means clustering approach to build a customer archetype segmentation across the store network. This became part of how real estate and merchandising leadership evaluate expansion decisions.
03 / Generative AI / Data Platform
GenAI Portfolio: Catalog Enrichment, Internal Search & Feedback Analysis
The Challenge
Three separate problems, one GenAI-shaped answer to each: product catalog data was messy and inconsistent at scale; internal SQL/code knowledge was locked away and undiscoverable by anyone who didn't already know where to look; and customer feedback was piling up unanalyzed because manual review didn't scale.
The Solution
An LLM-based enrichment pipeline that cleans and standardizes product title data - parallelized, with a multi-tier fallback for malformed model output and async writes so the pipeline doesn't stall on I/O. A hybrid BM25 + vector search system (RAG) that makes a large internal SQL/code repository queryable in natural language, running on a GCP-native stack with a dedicated vector database. An NLP pipeline for customer-feedback sentiment and theme extraction, with a constrained theme taxonomy enforced at the prompt layer so labels stay consistent run over run.
04 / Data Engineering
Automating a Manual Executive Reporting Process
The Challenge
A recurring executive-level analytics report was built on a slow, hard-to-maintain query pattern - expensive to run and risky to modify.
The Solution
Refactored it into a stored procedure: restructured an inefficient join pattern, pre-filtered the underlying data upfront, and added an audit table so date-range logic could be validated automatically instead of by hand.
05 / Personal Project
CueLens - Real-Time AI Interview Copilot
The Challenge
Real-time AI assistance during live conversation is a genuinely hard systems problem - audio input and manual triggers were racing each other and producing duplicate or ungrounded responses.
The Solution
A desktop copilot integrating Whisper speech-to-text with an LLM orchestration layer. I engineered a layered runtime prompt system (merging style rules and a hallucination guard by response level), fixed a dual-response race condition between the manual-trigger and live-audio paths, and improved answer continuity with a 2-turn sliding context window.
06 / Personal Project
Voice Diary - AI-Powered iOS Journaling App
The Challenge
Voice journaling apps typically force a choice between simplicity and intelligence - either a basic recorder, or a cloud-dependent app that doesn't respect data ownership.
The Solution
A local-first, Apple-native journaling app - SwiftData as the source of truth, optional CloudKit sync, iCloud Drive for audio - with a Whisper-via-Firebase transcription pipeline and semantic search over entries (SQLite vector storage, embeddings, cosine similarity) so users can have AI conversations grounded in their own diary history. Secured with Face ID/Touch ID and Firebase App Check.
Additional Work
Customer 360 Platform
problem: Customer identity was fragmented across a third-party data source and internal systems.
built: Identity resolution plus cohort/channel analytics (LTV, recency, category breadth, discount behavior).
Segment Analytics
problem: A key customer segment needed recurring, trustworthy tracking, and existing channel-classification logic had a bug.
built: Recurring cohort analysis, plus caught and fixed the classification bug.
Sales & Margin Scenario Planner (business case)
problem: Leadership wanted a new planning tool, but feasibility was unknown.
built: A phased roadmap and feasibility assessment - and caught a critical data-join gap before any engineering work started.
AI Tooling Adoption Business Case
problem: The data science team needed internal approval to use Claude AI, blocked on both value justification and legal/InfoSec review.
built: The proposal that secured it, covering ROI plus DPA/IP/vendor-risk requirements.
Foundational Analytics
problem: Recurring need for segmentation and reconciliation across systems.
built: Customer segmentation, cross-system discount data reconciliation, multi-channel identity bucketing, and new-store cohort analysis.
Career
Experience
Mar 2023 – Present
Katy, TX
Senior Data Scientist - Current at Academy Sports + Outdoors
- Built and own a production customer propensity model - feature engineering, calibration, and the full retraining/scoring pipeline behind it
- Architected a Customer 360 platform combining third-party identity resolution with cohort and channel analytics
- Refactored a slow executive reporting process into a maintainable stored procedure
- Built a generative AI portfolio - catalog data enrichment, internal RAG search, and customer-feedback NLP - on Vertex AI and LangChain
- Built forecasting and clustering models supporting real estate and merchandising strategy
Sept 2021 – Dec 2022
Chicago, IL
Data Analyst at Illinois Institute of Technology - Stuart School of Business
- Built interactive Power BI dashboards tracking academic KPIs for department leadership
- Wrote SQL against student records to surface enrollment and performance trends
- Mentored junior analysts on SQL and reporting practices
Feb 2019 – May 2020
Chennai, India
Research Data Analyst at SRM Institute of Science and Technology
- Led a team of four on qualitative research using AWS and statistical analysis in R
- Produced three peer-reviewed publications spanning image clustering, signal processing, and security
Nov 2018 – Jan 2019
Chennai, India
Data Analyst Intern at SLN Technologies
- Wrote SQL against multiple systems and built ETL workflows into data marts
- Built operational reporting infrastructure in Tableau, including logistics-focused reporting
Publications
Oct 2020
Study of the Clustering Algorithms for Hyperspectral Remote Sensing ImagesCompared DBSCAN, MiniBatch K-Means, and K-Means for classifying hyperspectral imaging data (Salinas Valley, CA dataset); K-Means achieved the best cluster parity (89.27% efficiency).
Apr 2020
Real-Time ECG Signal Analysis for At-Home Cardiac MonitoringIoT-based system classifying real-time ECG readings using Neurokit for feature extraction and CNN for classification, trained on Physionet Challenge 2017 data. 76–84% signal-quality accuracy, target cost under $20/unit.
Apr 2019
Securing EVMs Against Tampering with AES Encryption and HashingA tamper-detection method for electronic voting machines using AES encryption paired with multiple hashing algorithms, tested across various key sizes.
May 2020
Unbox the Brain for Dreams Using AIA concept piece exploring how AI and augmented reality could be applied to studying dream states.