I build the data and AI systems that take manual work out of operations — then measure what they actually saved.
Most recently I was AI Growth and Marketing at JustPaid (YC W23), where I owned the analytics stack end to end: the pipelines underneath it, the agents running on top of it, and the reporting 15+ leaders opened every morning. Before that, workforce and operations analytics for a 500-person centre, and reporting automation inside an enterprise SaaS platform.
What I'm good at: turning a messy, undefined business question into a working system — SQL and Python underneath, an interface someone non-technical will actually use on top, and a number attached to the outcome.
🎯 Open to: Business / Data Analyst, Analytics Engineering, and AI Automation Program roles — Seattle or remote.
📍 Seattle, WA · 🔗 LinkedIn · 🌐 Portfolio · ✉️ shrinijakummari@yahoo.com
| What it does | Why it mattered | |
|---|---|---|
| Agent Eval Harness | pass@k, judge panels with bias controls, a release gate that blocks CI | Found a third of test cases pass and fail the same test — invisible if you run each once |
| Automation Business Case | Models what deflection actually saves, not what volume × AHT implies | The naive case was 1.92x optimistic — $725K of savings that were never there |
| Workforce Shift Optimizer | Forecast → Erlang C → constraint solver → validated roster | Found a 21% cut in paid hours at identical service level, and a 27% coverage gap the spreadsheet had missed |
| JustPaid Analytics Platform | 4 platform APIs → BigQuery → Streamlit, scheduled on Cloud Run | Replaced manual reporting across 4 platforms and 3 spreadsheets — 40% fewer manual cycles |
| Job Agent | Scrapes roles, scores fit, tailors a résumé, emails a digest | An end-to-end pipeline with a human approval step, running unattended |
| AI Research Agents | Specialist agents that collaborate to produce a research report | Multi-agent orchestration with bounded tasks and structured handoffs |
| Ava | Voice-first desktop assistant — live audio, screen awareness | Built on the Gemini Live API for the Google Agent Challenge |
Data — SQL · Python (pandas) · BigQuery · Redshift · PostgreSQL · ETL & API integration · dbt-style modelling AI — Claude API · Claude Code · MCP servers · Gemini · LangChain · multi-agent orchestration · prompt engineering & eval Ship it — Google Cloud Run · Docker · Git · scheduled jobs · logging, alerting & cost tracking Show it — Streamlit · Tableau · QuickSight · Plotly · Excel (the unglamorous one that runs the world)
Certified: Claude Code in Action — Anthropic Academy, June 2026