Looking for a job

riya gaur

@riyagaur

Message

I evaluate AI systems and build data products that improve outcomes.

United States

Message

What I'm looking for

I'm looking to apply AI evaluation, product analytics, and data-product building skills to improve model quality, operational performance, and user outcomes.

I've evaluated 4,400+ LLM responses and agent trajectories across Claude, GPT-4, and Gemini, translating open-ended model behavior into measurable acceptance criteria, benchmarks, and failure taxonomies.

At AmpUp AI, I was the only analyst, owning product analytics across 650,000 events and building the PRD, data architecture, and entity-resolution pipeline for a 48-state operators database. I also cut LLM operating cost by 80% through prompt caching, batching, and parallel execution.

At Turing and Pareto AI, I've improved evaluation throughput while maintaining 96% QA acceptance, investigated model failures, and built Python validation checks that focus review time on meaningful model behavior.

Experience

Work history, roles, and key accomplishments

Current

AI Agent Web Specialist

Current

Turing

Jul 2026 - Present (4 months)

Evaluated 400+ multi-step agent trajectories across Claude, GPT-4, and Gemini at 96% QA acceptance, scoring full action sequences. Diagnosed 110 failed trajectories, escalated 20 invalid tasks, and raised throughput from 45 to 65-70 trajectories per week.

LLM Evaluation

Agent Trajectory Analysis

Claude GPT 4 Gemini Prompt Engineering

Failure Analysis

AI Data Analyst, Model Evaluation

Pareto AI

Nov 2025 - Jun 2026 (7 months)

Evaluated 4,000 LLM responses at 96% QA acceptance and 94% gold-task agreement, designing 300+ structured test scenarios. Built a failure taxonomy, wrote 10 automated validation checks, and analyzed 900 lower-scoring responses to identify performance gaps.

LLM Evaluation Python Pandas SQL Test Scenario Design

Failure Taxonomy

Rubric Alignment

Product Data Analyst

AmpUp AI

Aug 2024 - Dec 2025 (1 year 4 months)

Owned analytics scoping through delivery across 650,000 product events and 20 accounts at an early-stage AI startup. Cut LLM operating cost by 80%, wrote the PRD for a 48-state data product, and improved activation and adoption metrics.