SRINIVASAN_RESUME.PDF · TRANSMISSION RECORD
► DOWNLOAD
IDENT · SS·08·15·05 · AUSTIN, TX · CS @ UT AUSTIN · EXP. MAY 2030

Sriram
Srinivasan

ML engineer building systems that turn messy data into decisions. CS student at UT Austin. Research internships at NeuroSpine Nexus (UT Southwestern) and AIEA Lab (UC Santa Cruz).

ACTIVE · 3 TRANSMISSIONS FREQ · ML / DATA SYSTEMS EXP. MAY 2030
► CONTACT ► GITHUB
RESEARCH EXPERIENCE
2 POSITIONS
ML RESEARCH INTERN
AIEA Lab @ UC Santa Cruz
SANTA CRUZ, CA · JUNE 2025 — AUG 2025
● COMPLETED
  • Designed experiments comparing LiDAR vs. camera sensor accuracy for autonomous navigation
  • Processed 3D point clouds from Waymo and nuScenes datasets under varying motion and lighting
  • Applied ADE metric and ML simulations to optimize LiDAR-camera fusion weighting
  • Demonstrated hybrid sensor configs outperform single-sensor setups on trajectory modeling
RESEARCH INTERN
NeuroSpine Nexus
UT SOUTHWESTERN · JUNE 2022 — MAY 2025
● COMPLETED
  • Assisted across 70+ projects with 40+ researchers spanning 15+ countries
  • Chart review, data collection, and manuscript drafting for academic publication
  • Univariate and multivariate statistical analyses in R and SPSS
  • Applied LLM-powered document processing to categorize scientific literature at scale
TX·001 · NBA PRESEASON PROJECTION · ACTIVE  ·  TX·002 · CUSTOMER CHURN PREDICTION · DEPLOYED  ·  TX·003 · LA TIENDITA MCKINNEY · SHIPPED  ·  AIEA LAB · UC SANTA CRUZ · COMPLETED  ·  NEUROSPINE NEXUS · UT SOUTHWESTERN · COMPLETED  ·  XGBOOST · SHAP · FASTAPI · STREAMLIT · PYTORCH · SKLEARN  ·  TX·001 · NBA PRESEASON PROJECTION · ACTIVE  ·  TX·002 · CUSTOMER CHURN PREDICTION · DEPLOYED  ·  TX·003 · LA TIENDITA MCKINNEY · SHIPPED  ·  AIEA LAB · UC SANTA CRUZ · COMPLETED  ·  NEUROSPINE NEXUS · UT SOUTHWESTERN · COMPLETED  ·  XGBOOST · SHAP · FASTAPI · STREAMLIT · PYTORCH · SKLEARN  · 
TRANSMISSIONS
3 PROJECTS
TX·001 · 2026 — ONGOING ● ACTIVE   FULL PIPELINE

NBA Preseason
Projection

CALLSIGN · HOOPS·ML·01 · SPORTS FORECASTING
PROBLEM
Predict preseason player performance from historical data without peeking at the future. Translate ensemble forecasts into lineup recommendations with honest confidence intervals.
METHODOLOGY
nba_api data collection with strict temporal guards. Pace-adjusted stats, role indicators, recency decay. Multi-output XGBoost with quantile regression for 10th/90th percentile intervals. SHAP per-stat attribution. FastAPI and Streamlit.
KEY FINDING
Random train/test splits inflate validation accuracy by up to 12% on this dataset relative to season-stratified splits. Future game stats leaking into training features are the primary source of variance inflation, not model complexity.
PYTHONXGBOOSTNBA_APISHAPFASTAPISTREAMLITSCIKIT-LEARNPANDAS
► GITHUB
TX·002 · 2026 ● DEPLOYED   EXPLAINABILITY

Customer Churn
Prediction

CALLSIGN · CHURN·XGB·02 · TELCO / BUSINESS ANALYTICS
PROBLEM
7,043 telecom customers. 20 features. Which ones leave and why? Build a model that predicts churn and explains each customer's risk in terms a retention team can act on.
METHODOLOGY
XGBoost tuned for recall over precision: 0.773 recall, 0.841 ROC AUC. SHAP TreeExplainer for per-instance feature attribution. FastAPI inference layer with Streamlit dashboard for analyst-facing output.
KEY FINDING
Contract tenure and monthly charges account for over 60% of mean absolute SHAP contribution across the test set. High-cardinality service features contributed noise rather than signal and were dropped after ablation.
PYTHONXGBOOSTSHAPFASTAPISTREAMLITSCIKIT-LEARNMATPLOTLIBPANDAS
► GITHUB
TX·003 · JUNE 2025 — OCT 2025 ● SHIPPED   SOCIAL IMPACT

La Tiendita
McKinney

CALLSIGN · TIENDITA·01 · NON-PROFIT / COMMUNITY
PROBLEM
A non-profit serving low-income Latino families in McKinney. Career guidance running on paper and gut instinct. Limited budget, no technical staff. 15+ students with no structured pathway to professional opportunities.
APPROACH
OpenAI API integration for personalized career Q&A, prompt-engineered for a bilingual student context. Lightweight Streamlit frontend designed around non-technical users. Zero-maintenance deployment with no backend infrastructure overhead.
KEY FINDING
Prompt structure had more impact on output quality than model selection. Few-shot examples grounded in the local job market reduced hallucinated credential requirements by a measurable margin in manual eval.
PYTHONOPENAI APINLPSTREAMLITPANDAS
► GITHUB
SIGNAL SPECS
TECHNICAL STACK
ML / AI
PyTorch · XGBoost · scikit-learn · SHAP · quantile regression · LiDAR fusion
DEPLOYMENT
FastAPI · Streamlit · Vercel · Streamlit Community Cloud
DATA
pandas · numpy · matplotlib · nba_api · Waymo · nuScenes
LANGUAGES
Python · Java · R · SPSS · SQL
RESEARCH
Univariate / multivariate stats · clinical data · LLM document processing · ADE / trajectory modeling
EDUCATION
B.S. Computer Science · UT Austin · Expected May 2030
OPEN CHANNEL
Looking for internships · Summer 2027 and beyond
ML engineering · data systems · research roles
► SEND MESSAGE