Vyas Kumar
Data science and AI. Based in Salt Lake City.
Background
I build the systems that turn data into decisions: pipelines, models, and the infrastructure they run on.
At REDX I own the marketing team's infrastructure. That means the full lifecycle of a lead: capture, attribution, routing into the sales flow, and re-engagement once it goes quiet. I designed and built that system from the ground up, so every action a lead takes is tracked and attributable.
That work does not stay inside marketing. The same warehouse feeds Product, Sales, and Customer Success, so a lead source recorded at capture is still readable months later next to what the account went on to do. The reporting layer on top of it is what executives read instead of asking someone to pull numbers by hand.
The projects I gravitate toward involve causal estimation and statistical analysis, machine learning, retrieval systems over large text corpora, and agent building.
Experience
REDX
Jul 2024 – PresentMarketing Analyst
- Own end-to-end email marketing for 100K+ contacts via customer data platform
- Own the segmentation rules engine producing mutually exclusive audiences, with engagement tiers read from human opens rather than the machine-inflated gross count
- Architect the lead pipeline from customer data platform into the sales platform: 100% sync rate, one-minute median, 99.8% inside five minutes
- Own the BigQuery reporting layer behind executive dashboards, eliminating manual data pulls entirely
- Work with Sales and Product on lead routing rules and the attribution both teams report against
Selah Digital
Sep 2023 – Jan 2024Salesforce Business Analyst Intern
- Set up campaign architecture in Salesforce Marketing Cloud and integrated it with Health Cloud
Novac Technology Solutions
Jan 2019 – Mar 2019IT Intern
- Analyzed ticket trends and performed root cause analysis, reducing ticket traffic by 5%
- Maintained cloud infrastructure
- Reworked the invoicing workflow to remove duplicate manual entry
Skills
Machine Learning & Statistics
- Causal Inference
- Propensity Score Modeling
- Doubly Robust Estimation
- Sensitivity Analysis
- Time Series Forecasting
- Probabilistic Forecasting
- Uncertainty Quantification
- Rolling-Origin Backtesting
- Predictive Modeling
- Model Interpretability
- Experiment Tracking
- A/B Testing
AI Engineering
- Retrieval-Augmented Generation
- Hybrid Retrieval
- Vector Embedding & Indexing
- Structure-Aware Chunking
- Parent-Child Retrieval
- Reranking
- Semantic Search
- Metadata Filtering
- LLM Enrichment & Classification
- Query Routing
- Citation Grounding
- MCP Servers
- Multi-Agent Orchestration
Data Engineering
- Warehouse Schema Design
- ETL Pipeline Development
- Reverse ETL
- Change Data Capture
- Deduplication & Canonical Modeling
- Identity Resolution
- Pipeline Orchestration
- Out-of-Core SQL Processing
- API Integration
- Data Quality Monitoring
- Least-Privilege Access Design
- System Architecture
Analytics & Reporting
- Dashboard Design
- Self-Service Analytics
- Executive Reporting
- Metric Instrumentation
- Funnel Instrumentation
- Lead Attribution
- Data Visualization
Marketing Systems
- Lifecycle Automation
- Audience Segmentation
- Lead Capture & Routing
- Lead Scoring & Qualification
- Deliverability Monitoring
- List Validation
- Campaign Measurement
- CRM Data Integration
How I work
Systems over scripts
One-off solutions are a tax you pay later. I would rather spend the extra time up front than rebuild the same thing six months down the line.
Evidence beats opinion
Anyone can tell a good story about a number. I care whether it held up once we tested it properly.
Validate against known answers
If a pipeline lands outside an established range, I assume I broke something and go looking. Build the check before you trust the result.
Documentation is part of the deliverable
If the next person cannot operate it, it is not done.