Akash Sathish
Portrait of Akash Sathish

Akash Sathish

Computer Science and Data Science, University of North Carolina at Chapel Hill

Computer Science and Data Science at UNC, minor in Artificial Intelligence.
Incoming Software @ Fidelity Investments. Previous AI Engineer at Snapback Sports.

Experience

Snapback Sports

Jun 2026 – Sep 2026
AI EngineerNew York, NY
  • Built the LLM pipeline that generates trivia questions, with schema validation and prompt caching. Accuracy held above 98% on an 800+ question evaluation set.
  • Reduced incorrect questions by grounding generation in sourced sports data (RAG) and adding an LLM-as-judge review step before anything was published.
  • Shipped new features and game modes in the Snapback Trivia app (100K+ downloads) using React Native.
  • Shipped Snapback Experiences, a gameday planning web platform that handled more than 1,000 concurrent users.
  • Built an offline-first iOS app with multi-device sync and session authentication, released to TestFlight for internal beta.
PythonLLMsRAGReact NativeSwift

Intelitrade

Jan 2026 – May 2026
Software EngineerChapel Hill, NC
  • Developed an LLM sentiment analysis system using the OpenAI API and LangChain, validated against human ratings on more than 400 articles.
  • Built an internal AI agent for developer workflows with tool-scoped permissions and human approval steps. It was adopted across the team.
OpenAI APILangChainPython

UNC Klotsa Group

Dec 2025 – Jan 2026
Research Software DeveloperChapel Hill, NC
  • Built a faculty profile system in Next.js and PostgreSQL for more than 40 faculty, replacing a spreadsheet maintained by hand.
  • Connected profile edits to the matching engine in real time, so matches for 90+ clinician and faculty pairs updated immediately.
Next.jsPostgreSQL

Projects

MailSentry

Available on the Mac App Store

A macOS app that lets AI assistants work with your email without giving them access to your actual mail account. It runs entirely on your Mac and does not use any external servers.

  • Apple MailKit extension and a sandboxed launchd helper that give assistants a read-only IMAP copy of your mail.
  • Limits that copy to mail from your primary address, cleaned by a MIME parser and several detectors for hidden text.
  • Any action an assistant takes goes through an MCP server with rate limiting and requires your approval.

Arceo

In development

A platform that estimates what an AI agent will cost to run and what the worst outcome of its actions could be, before it is deployed.

  • Three-tier cost forecasting engine that predicts an agent's LLM spend before deployment, within 7% of actual spend.
  • Scoring engine that identifies the most damaging thing an agent could do based on the tools it has access to.
  • Found a sequence in OpenAI's six-agent customer service demo that could issue refunds without approval.
  • GitHub Actions workflow that checks agent safety on each pull request and blocks the merge when risk is too high.

BackBuddy

March 2026

A trading assistant for Kalshi sports markets that follows several NCAA games at once.

  • Integrated Kalshi's REST API with a FastAPI backend for automated trade execution.
  • Live profit and loss tracking, with custom React hooks for alerts.
  • Cash-out warnings tied to the current state of each game.

Education

University of North Carolina at Chapel Hill

Expected May 2028
DegreesB.S. Computer Science, B.S. Data Science
MinorArtificial Intelligence
CourseworkData Structures and Algorithms, Computer Programming, Systems Fundamentals, Discrete Mathematics
HonorsDean's List, Fidelity Hackathon Winner, Eagle Scout

Technical skills

AreaSkills
LanguagesPython, Swift, SwiftUI, Java, SQL, TypeScript, JavaScript, C
Frameworks and toolsFastAPI, Next.js, React Native, PostgreSQL, Cloudflare Workers, Xcode, Swift Testing

Activities