At Elevate Credit I owned the model inventory for 10+ active credit models, automated performance tracking that cut execution time by 95%, and replaced a retiring SaaS platform with a fact-and-dimension database. At Apex Informatics I developed a centralized Power BI dashboard tracking 20+ operational KPIs and implemented an AI-powered text-to-SQL solution with automated regression testing. I work in Python, SQL, Power BI, and Snowflake.

const viraj: Engineer = {
  name: "Viraj Shetty",
  role: "Credit Risk Analyst | Data Scientist",
  company: "Apex Informatics",
  location: "Dallas, TX",
  yearsShipping: 7,
  languages: ["R", "SQL", "Python", "Dart", "Jupyter Notebook", "Makefile"],
  stack: ["PyTorch", "Keras", "Scikit-learn", "TensorFlow", "XGBoost", "LangGraph", "Power BI", "Snowflake"],
  career: [
    { at: "Apex Informatics", role: "Operational Analyst", when: "Sep 2026 – Present" },
    { at: "Elevate Credit", role: "Data Scientist II", when: "Aug 2023 – Jul 2026" },
    { at: "Loopback Analytics", role: "Data Analyst Intern", when: "Jun 2022 – Dec 2022" },
    { at: "mUni Campus", role: "Data Analyst", when: "Dec 2019 – May 2021" },
  ],
  education: "MS in Information Technology Management (Business Analytics Specialization); relevant coursework: Machine Learning, Advanced Statistics for Data Science, Database Foundations, Big Data, The University of Texas at Dallas",
};

September 2026 – Present

Developed a centralized Power BI dashboard tracking 20+ operational KPIs across project delivery, resource utilization, client performance, and financial operations, reducing reporting effort by 35%.

Streamlined SQL and ETL workflows to consolidate data from 5+ internal sources, standardizing operational reporting and improving data accessibility.

Implemented an AI-powered text-to-SQL solution with automated regression testing through Hermes.

August 2023 – July 2026

Standardized the model development lifecycle and replaced email-based reviews with structured leadership alignment meetings, reducing turnaround time by 75%.

Automated variable stability (PSI) and performance (AUC, KS, lift) tracking for 10 active models via parameterized Jupyter notebooks and an engineered Snowflake-to-Power BI pipeline, reducing execution time by 95%.

Built a fact-and-dimension database architecture to replace a retiring Tableau/SaaS reporting system.

June 2022 – December 2022

Used Snowflake to write stored procedures to optimize execution time and space by 25%; optimized Power BI data structure to reduce report load time by 30%.

Identified 5 key business metrics and conducted hypothesis testing on patient data, improving treatment success rate by 20%.

Utilized PySpark to manage and update NLP pipelines for EHR data and used Databricks and R to perform ETL on socio-economic data from Census API, reducing null values by 30%.

December 2019 – May 2021

Produced a Python-based YouTube scraper that scraped 10K links and titles, and implemented Random Forest and KNN machine learning algorithms achieving 92% precision and 89% recall for student performance prediction.

Performed ETL using pandas to store data with optimized SQL scripts, increasing access speed by 67%.

Collected and integrated data from multiple sources using Apache Airflow and AWS S3, reducing manual effort by 50%.

email · github

• purgedcv — Purged, embargoed walk-forward cross-validation for time series that reports how much of a model's backtested skill was leakage, using a known-zero-signal…
• sliceguard — A fairness slice auditor using Wilson intervals, Benjamini-Hochberg correction and minimum detectable effect so tiny slices stop producing false alarms.
• quantfrontier — A quantisation quality/latency frontier measured on downstream tasks and output divergence rather than perplexity, using real integer quantisation on a small…
• picklescope — Static scanner for ML model files that inspects pickle, PyTorch, Keras, safetensors, GGUF, and ONNX without ever deserializing them, with text, JSON, and SARIF…
• agentguard — Capability-based permission engine for LLM agent tool calls, with YAML policies, argument constraints, taint tracking, human approvals, and a hash-chained…
• mcpgate — A Python security gateway for the Model Context Protocol that fronts many MCP servers with OAuth bearer validation, per-client tool RBAC, rate limits, audit…
• fieldvoice — An offline speech-to-report pipeline evaluated on entity error rate, with catalogue-constrained decoding that snaps part numbers and fault codes.
• receiptlens — A privacy-first Flutter receipt scanner: on-device OCR and a TFLite line-item classifier with fully local encrypted storage and CSV/QuickBooks export, with no…
• tamperlog — A tamper-evident audit log that adds Merkle checkpoints with externally anchored roots, so it detects re-chained truncation that plain hash chaining silently…
• yaraci — A CI harness for YARA rule quality that lints rules, proves they fire on fixtures, scans goodware for false positives, profiles scan cost and diffs rule…