Radha Shah

About Radha Shah

At Rightfoot, I bring my unicorn-scale experience (Nextdoor, Skillz) to lead technical strategy in a pre-PMF company\'s 0-1 journey. Most crucially, I led the team to build the company\'s first public API. I also drove $1.5M in cost savings, created an extensible transaction processing system covering major banks, including Chase, and developed BalanceIQ - a predictive balance calculation system that won the company hackathon. Through it all, I\'ve scaled systems 2X and improved operational efficiency from hours to seconds while fostering a culture of engineering excellence.At Nextdoor, I was a full-stack engineer working in Python and React to create features for the real estate section.At Skillz, I primarily worked on creating Java microservices from a Groovy on Grails monolith. This involves planning, designing, coding, testing and deploying endpoints using a host of technologies such as Java, Spring Boot, AWS, MySQL etc. At Citi, I played a variety of software engineering roles at a financial services firm. Initially, I worked on the trading floor making VBA scripts for the traders and salespeople. I then supported the kdb processes for a Delta One Derivatives application, and created kdb-related features in C# and Java for various trading applications. Next, I helped code front-end features for a FX trading platform, before finding my home in the back-end - creating server-side features in various microservices related to currency pair trading.

Experience

1. Staff Software Engineer

   Rightfoot

   Jan 2025 — Present · San Francisco, CA, US

   Architected a production-grade agentic AI system featuring fine-tuned models, LLM-as-judges, task evaluators, safety guardrails, and observability, expanding the platform toward universal coverage.• Increased coverage 20×, enabling a clear path toward future integrations• Developed fine-tuned OpenAI models with LangFuse traces to generate high-quality training data• Curated evaluation and golden datasets and built Kiln-based task-performance evaluators to benchmark fine-tuned vs. frontier models• Used LangFuse trace-level observability to support logging, debugging, and iterative refinement• Introduced safety guardrails, including risk detection heuristics, to ensure the agent operated within defined boundaries.* Designed prompt-engineered LLM-as-judge to score task outcomes per risk, generate confusion matrices, and systematically eliminate false negatives.• Redesigned the integration architecture to eliminate ~90% of per-integration code, dramatically simplifying large-scale onboarding

Education

• Rutgers University

  B.S. in Computer Science, B.A. in Psychology

Skills

• Spring Framework
• Os X
• Jquery
• Html
• Time Management
• Unix
• Kdb+
• Css
• Wpf
• Javascript
• Visual Basic
• Maven
• Bash
• Eclipse
• C#
• Shell Scripting
• Php
• Databases
• Sharepoint
• Apache
• Vba
• Vmware
• Tcp/Ip
• Automation
• Java
• Servers
• C
• Kdb
• Python
• Regression Testing
• Mysql
• Ubuntu
• Git