hi, I’m Viraj.
an AI/ML engineer & data scientist,
building AI tools and trustworthy decision systems.
a little room for curiosity.
Data, decisions, and a little room to think.
a small collection
Things I’ve made.
Applied AI, financial analytics, and models that inform decisions.
Where I’ve worked
the person behind the patterns
Curious by default.
I’m Viraj, an AI/ML Engineer at Apex Informatics in Dallas. I build natural-language-to-SQL tools, AI-assisted reporting, and automated data workflows. My background also spans credit-risk modeling, model monitoring, and governance.
I care about turning complex data into decisions people can understand — with review, validation, and clear assumptions built into the workflow.
- Implemented an AI-powered text-to-SQL application with automated Hermes regression testing.
- Integrated project, timesheet, client, and financial data from 5+ internal sources using SQL and ETL pipelines.
- Configured Python/SQL agents for recurring reporting and data-quality validation, reducing reporting turnaround by 4 hours.
- Built a Streamlit report with natural-language explanations of resource utilization, SLA performance, and budget variance.
- Engineered Clay, HubSpot CRM, and n8n workflows for lead enrichment, segmentation, CRM synchronization, and outreach.
- Standardized model-development reviews and leadership alignment, reducing turnaround time by 75%.
- Presented model-performance reviews and helped retire 2 underperforming models in one review.
- Drove a LangGraph RAG-based underwriting-document review tool, reducing review time by 90% per document.
- Managed the lifecycle of 10–12 active models and delivered status updates within 24 hours of a model change.
- Coordinated third-party model validation and fair-lending reviews.
- Built an XGBoost credit-risk model with reject inference, reducing projected delinquent loans by $1M.
- Automated PSI, AUC, KS, and lift monitoring for 10 active models through Jupyter, Snowflake, and Power BI, reducing execution time by 95%.
- Built charge-off benchmarking, roll-rate, and re-parameterization views; the résumé reports a 60% reduction in the delinquent population.
- Refined underwriting lookup queries with Claude, reducing manual investigation time by 85%.
- Optimized Snowflake stored procedures by 25% and Power BI report-load time by 30%.
- Identified 5 business metrics and ran hypothesis tests on patient data, with a résumé-reported 20% improvement in treatment success rate.
- Used PySpark for EHR NLP pipelines and Databricks/R for Census-data ETL, reducing null values by 30%.
The toolkit
Python · SQL · R · pandas · NumPy · PyTorch · TensorFlow · Keras · scikit-learn · XGBoost · LangGraph · REST APIs
Power BI · Tableau · Snowflake · Databricks · Excel
Docker · Azure · AWS · Git · Linux
Claude · GitHub Copilot · ChatGPT · Perplexity · Databricks Genie AI
Education
The University of Texas at Dallas
Experience and outcomes are reported in my current résumé.
The longer story, on paper ↗a little more human
Beyond the models,
I’m just Viraj.
I’m into music, strategy games, and questions that send me down a rabbit hole.
Whether it’s AI, a project idea, or a song recommendation — I’d love to hear from you.
Dallas, Texas · open to US-remote work