Data Analyst, Data Scientist, Data Engineer, AI Engineer, AI / ML Engineer, Analytics Engineer
Pavan Mallipudi · Tempe, AZ
I work end to end: the pipelines that move the data, the models and retrieval that reason over it, and the dashboards and APIs that put it in front of people. The part I care about most is the handoff, where a number becomes a decision or a model becomes something you can actually call.
B.S. Computer Science · Minor in Data Science · Magna Cum Laude
Churn across 23M KKBox billing transactions and 410M member-days. Manual renewers churn at 20.9% against 2.8% on auto-renew, and a gradient-boosting model scores 0.944 AUC on held-out 2017 decisions, its top risk decile capturing 82% of churners against 68% for a rules baseline. dbt and DuckDB behind a Next.js dashboard.
Approve or decline, scored on profit and fairness rather than accuracy alone. XGBoost over 618,584 LendingClub loans with time-based validation; declining the riskiest 12% lifts 2015 profit by 13.8%. A fairness monitor reports who that costs.
Ask questions about your own PDFs and get answers traceable to the exact chunk. Runs fully local: MiniLM embeddings, Chroma, and Qwen2.5-1.5B behind FastAPI. Retrieval takes the benchmark set from 1/8 to 6/8 correct.
Double-entry bookkeeping for freelancers. Bank CSV import with duplicate detection, rules-then-LLM categorization, invoices and AR aging, and statements built straight from the ledger. FastAPI, PostgreSQL and React, multi-tenant, 153 backend tests.
Identifies 38 leaf diseases across 14 crops from a single photo. EfficientNet-B0 transfer learning on PlantVillage, 98.21% test accuracy from a 4.06M-parameter model, served through a Streamlit app that flags low confidence.
CMS Medicare readmissions end to end: live ingestion, cleaning and QA, statistical testing, and ownership and star-rating enrichment, in a deployed React dashboard for exploring excess readmission patterns.
297 real patients as an orbitable 3D point cloud, with PCA, k-means clustering, and a logistic-regression risk model at ROC-AUC 0.906 under stratified 10-fold cross-validation.
A Databricks lakehouse in Bronze, Silver and Gold layers. PySpark and Delta Lake MERGE for incremental loads, schema harmonization and data-quality validation from S3, governed by Unity Catalog.
I travel whenever I get the chance, and the trips I like most are the ones that involve a trailhead: national parks, long climbs, and the kind of cold that makes you check the forecast twice. Same instinct as the work, really: pick a hard problem, break it into switchbacks, keep moving.
B.S. Computer Science · Data Science minor · Magna Cum Laude · Arizona State University, May 2026. Open to data analyst, data engineer, data scientist, and AI / ML engineer roles.
pvmmallipudi@gmail.com · Tempe, Arizona