~$viraj.shetty
Menu

Data Scientist & Analyst

Credit risk and operational analytics, made legible.

I build Power BI reporting, SQL and ETL workflows, and automated model monitoring. I am an Operational Analyst at Apex Informatics.

View projects Résumé Get in touch

01 · About

Between the number and the system that produces it.

I build the reporting, pipelines and monitoring that let a team trust the numbers it reads.

My work sits at the join between analytics and data engineering: a question from the business has to travel through SQL, a warehouse and a dashboard without losing its meaning on the way.

Outside work I write small Python tools that test one claim each, from leakage-aware validation to fairness audits and tamper-evident logs.

Focus
Data analytics & credit risk
Base
Dallas, TX
Looking at
Analytics, reporting and credit risk work

02 · Experience

Roles, in order.

  1. September 2026 — Present

    Operational Analyst

    Apex Informatics

    Operational analytics and reporting, plus GTM and revenue operations automation with Clay, HubSpot CRM and n8n for lead enrichment, prospect segmentation and CRM synchronization.

    • 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 project, timesheet, client and financial data from 5+ internal sources into standardized operational reporting.
    • Implemented an AI-powered text-to-SQL solution so stakeholders can retrieve operational insights in natural language, with automated regression testing through Hermes to validate query accuracy.
    • Analyzed utilization, capacity, SLA performance and budget variance across 15+ client engagements using Power BI, DAX and Power Query.
    • Automated recurring reporting and data quality validation with Python, SQL, Power Query and Excel, reducing reporting turnaround by 4 hours.
    • Power BI
    • DAX
    • Power Query
    • SQL
    • ETL
    • Python
    • Excel
    • Hermes
    • Clay
    • HubSpot CRM
    • n8n

  2. August 2023 — July 2026

    In progress

    Data Scientist II

    Elevate Credit

    Owned the inventory lifecycle of 10-12 active models, with status updates to internal stakeholders and bank partners within 24 hours of any model change. Coordinated third-party validation against SR-26-2 and OCC-11-12 and facilitated fair lending reviews.

    • Automated variable stability (PSI) and performance (AUC, KS, lift) tracking for 10 active models via parameterized Jupyter notebooks and a Snowflake-to-Power BI pipeline, reducing execution time by 95%.
    • Designed ad-hoc views for charge-off benchmarking, roll-rate analysis and re-parameterization comparisons, driving pre-rollout lift analysis that reduced the delinquent population by 60%.
    • Enhanced an automated lookup query that retrieves underwriting data by application ID on API-flagged fund/no-fund conflicts, using Claude for query refinement and cutting manual investigation time by 85%.
    • Built a fact-and-dimension database architecture from scratch to replace a retiring Tableau/SaaS reporting system, with a scalable schema for current and future credit risk and response models.
    • Standardized the model development lifecycle, replacing email-based reviews with structured leadership alignment meetings and reducing turnaround time by 75%.
    • Snowflake
    • Power BI
    • SQL
    • Jupyter
    • Tableau
    • Claude
    • Copilot

  3. June 2022 — December 2022

    Data Analyst Intern

    Loopback Analytics

    Analytics and data engineering internship on healthcare data, covering Snowflake, Power BI, PySpark and Databricks.

    • Tracked client feasibility requirements in JIRA and wrote Snowflake stored procedures that optimized execution time and space by 25%.
    • Optimized the Power BI data structure to reduce report load time by 30%.
    • Identified 5 key business metrics with a cross-functional team and ran hypothesis testing on patient data, improving treatment success rate by 20%.
    • Used PySpark to manage and update NLP pipelines for EHR data, and Databricks and R for cleaning and ETL on Census API socio-economic data, reducing missing values by 30%.
    • Snowflake
    • Power BI
    • PySpark
    • Databricks
    • R
    • JIRA
    • NLP
    • ETL
    • Census API

  4. December 2019 — May 2021

    Data Analyst

    mUni Campus · Mumbai, IN

    Data integration, root cause analysis and predictive modeling on academic performance and lecture data.

    • Integrated data from multiple sources using Apache Airflow and AWS S3 and redesigned API services for greater scalability, reducing manual effort by 50%.
    • Implemented Random Forest and KNN models for student performance prediction, achieving an average precision of 92% and recall of 89%.
    • Performed ETL with pandas and optimized SQL scripts, increasing data access speed by 67%.
    • Ran root cause analysis with regression and correlation analysis to identify factors affecting test scores.
    • Built a Python YouTube scraper that collected 10K links and titles for a specific keyword.
    • Python
    • pandas
    • SQL
    • Apache Airflow
    • AWS S3
    • Random Forest
    • KNN
    • ETL

03 · Projects

Small tools that test one claim each.

Each project answers a question about trust: whether a number holds up, a model is fair, a record was changed, or a rule does what it says.

QueryCraft: a question becomes SQL, is validated, then runs read-only question SQL draft validate SELECT month, SUM(rev) FROM orders GROUP BY month results

Text-to-SQL workbench

QueryCraft

A local-first natural-language-to-SQL workbench. It generates schema-grounded SQL, lets you review and edit it, and runs only validated read-only SQLite queries. It has a schema explorer, CSV export and a no-key demo mode.

  • SQLGlot validation pipeline with user-approved, read-only execution, query deadlines and row limits
  • Verified through 92 automated tests and browser regression checks
  • Python
  • Flask
  • SQLite
  • SQLGlot
  • JavaScript

Source on GitHub

Olist sales report: KPI tiles, a monthly bar chart and a map, in Brazilian flag colours revenue · map · customers revenue orders AOV customers revenue by month map

BI reporting

Brazilian Olist E-Commerce Reporting

A Power BI report on the Brazilian Olist e-commerce dataset with revenue, map and customer analytics views, built to showcase Power BI skills. It uses a custom theme matching the colors of the Brazilian flag.

  • Measures for customer counts, total revenue, total orders, average order value and average orders per customer
  • Revenue, map and customer analytics views
  • Power BI
  • SQL
  • DAX

Source on GitHub

Two credit series over time: revolving and nonrevolving balances diverge revolving nonrevolving 2000 1943 2026

Financial data analysis

US Consumer Credit Analysis

An analysis of the Federal Reserve G.19 Consumer Credit release (Jan 1943 to Jun 2026) comparing revolving and nonrevolving credit. It ships as a notebook with 7 charts and written findings, and as an interactive Streamlit dashboard.

  • Year-over-year growth of the two series correlates at only 0.16
  • Student loans grew 286% since 2006 and are now 49% of nonrevolving credit
  • Python
  • Jupyter
  • Streamlit

Source on GitHub

Forest plot of slice gaps: one interval clears zero, the rest stay indeterminate accuracy gap, interval per slice region, age, channel age, channel region, prior channel green = disparity survives correction

Fairness auditing

sliceguard

A fairness slice auditor with Wilson and Newcombe intervals on every slice, Benjamini-Hochberg correction across all slices tested, and a minimum detectable effect. Underpowered slices are marked INDETERMINATE.

  • Ranks by the lower bound of the gap interval instead of the raw point estimate
  • Demo audits 6,000 synthetic underwriting-model evaluations with two planted 9-point disparities
  • Python
  • Statistics
  • Wilson intervals

Source on GitHub

Walk-forward folds with a purge gap and embargo around each test block train purge and embargo test

Time-series validation

purgedcv

Purged, embargoed walk-forward cross-validation for time series, plus a harness that measures how much of a backtest's reported skill is leakage. It scores one model under four protocols on synthetic data where the true skill is known, including a zero-signal control.

  • On a zero-signal series, naive k-fold overstated accuracy by 8.35 percentage points
  • The label horizon is mandatory, so overlap leakage cannot be reintroduced by default
  • Python
  • Time series
  • Statistics

Source on GitHub

Hash-chained log with a checkpoint held in a separate witness store, and a truncated tail checkpointMerkle root witness storekept apart entry entry entry entry lost each entry hashes the one before it

Data integrity

tamperlog

A hash-chained audit log with Merkle checkpoints whose roots sit in a separate witness store. It catches re-chained truncation and edits that plain hash chaining accepts.

  • On a 20,000-entry log with a checkpoint every 1,000 entries, chain-only checks missed 3 of 7 tamper classes; with checkpoints all 7 were detected
  • Inclusion proofs for any entry, with domain-separated leaf and node hashes
  • Python
  • Merkle trees
  • SHA-256

Source on GitHub

Also built

  • Evaluation metrics

    fieldvoice

    Offline speech-to-report pipeline scored on entity error rate, not word error rate.

  • Model evaluation

    quantfrontier

    Quantisation quality against latency and memory, measured per task rather than by perplexity.

  • Quality testing

    yaraci

    CI harness for YARA rule quality: lint, fixture tests, false-positive scans and one exit code.

  • Static scanner

    picklescope

    Static scanner for ML model files that never deserializes them.

  • Permission engine

    agentguard

    Permission engine for LLM agent tool calls: default-deny policies, human approval and an audit log.

  • Security gateway

    mcpgate

    Security gateway for Model Context Protocol servers: bearer validation, per-client tool access and rate limits.

  • Flutter app

    receiptlens

    Privacy-first receipt scanner: on-device OCR, encrypted local storage and CSV export.

Other repositories

04 · Stack

The toolbox, set out as tables.

Tools grouped by how I use them, each listed like a parts table.

Programming & Data

table 01
  1. 01SQL
  2. 02Python
  3. 03R
  4. 04PostgreSQL
  5. 05Pandas
  6. 06NumPy
  7. 07Matplotlib
  8. 08Linux

BI & Data Platforms

table 02
  1. 01Power BI
  2. 02DAX
  3. 03Power Query
  4. 04Snowflake
  5. 05Tableau
  6. 06Microsoft Excel
  7. 07VBA
  8. 08Index-Match

ML & Tooling

table 03
  1. 01Scikit-learn
  2. 02XGBoost
  3. 03PyTorch
  4. 04TensorFlow
  5. 05Keras
  6. 06LangGraph
  7. 07Git
  8. 08VSCode

Cloud

table 04
  1. 01Docker
  2. 02GCP BigQuery
  3. 03Azure
  4. 04AWS S3
  5. 05AWS Redshift
  6. 06AWS Athena
  7. 07AWS Glue
  8. 08Databricks

LLM & GTM

table 05
  1. 01Claude
  2. 02GitHub Copilot
  3. 03ChatGPT
  4. 04Perplexity
  5. 05Databricks Genie
  6. 06Clay
  7. 07HubSpot
  8. 08n8n

05 · Education

Education.

August 2021 — May 2023

The University of Texas at Dallas

MS · Information Technology Management (Business Analytics Specialization)

Relevant coursework: Machine Learning, Advanced Statistics for Data Science, Database Foundations, Big Data

June 2016 — November 2020 · Mumbai, India

The University of Mumbai

BE · Computer Engineering

Relevant coursework: Analysis of Algorithms, Cloud Computing, Natural Language Processing, Data Warehousing

06 · Contact

Get in touch.

Interested in analytics, reporting or credit risk work? Send me a note and I'll get back to you.

virajvshetty47@gmail.com