Eshaan Punalekar — Irvine, CA

I build software thatactually works.

Computer Science @ UC Irvine, focused on Intelligent Systems, with a minor in Innovation & Entrepreneurship. I like building things that solve niche problems.

View ProjectsView Resume

01//about

Right now.

A snapshot of where my attention is, rather than everything I've done. Updated July 2026.

Working on

  • Interning at CliftonLarsonAllen on the digital team, building machine learning and data science tooling.
  • Actively building this site — currently the shader background and the recruiter-mode document system.
  • MarketEdge — extending the calibration scoring so the engine's own accuracy is measurable over a longer window, not just per-event.
  • Getting ReturnTrack from a shipped web app to an app store release.

Learning

  • Data pipeline orchestration in earnest — retries, backfills, and what it takes for a scheduled job to fail loudly instead of silently.
  • Practical evaluation for LLM-backed features: how to tell whether an extraction step actually got better or just changed.

Recently

  1. ProjectJuly 2026

    ReturnTrack

    Never miss a return window again

    Receipt-tracking app using Claude's vision API to extract line items and compute return and warranty deadlines against a self-researched policy database covering 70+ policies across 50 retailers.

  2. WorkJune 2026 — Present

    Data and Automation Intern

    CliftonLarsonAllen

    Building approval workflows, prompt-versioning tooling, and connectors for an internal AI platform that turns client meeting transcripts into structured automation opportunities.

  3. ProjectJune 2026 — Present

    MarketEdge

    In progress

    A prediction-market divergence engine — surfaces pricing gaps across markets for the same underlying event, and grades its own calibration over time.

See full journey →

02//stack

What I work with.

Languages

PythonTypeScriptSQLJavaScriptC++JavaHTMLCSSSwift

Web / Full-stack

Next.jsFastAPIPostgreSQLTailwindPrismaClerk

ML / AI

MCPPyTorchLangChainAzure OpenAIAnthropic APIPandasPrompt engineeringAPI integrationscikit-learnComputer visionAI model trainingOpenCVData labeling

Cloud / Tools

DatabricksPostmanDockerGitAzure FoundryPower AppsPower Automate

03//projects

Selected work.

Scout

Full-stack multi-sport recruiting platform

Shipped — not yet available to end users

Connects athletes with college programs. Began as a soccer-only recruiting tool and expanded into a multi-sport platform.

  • Next.js
  • TypeScript
  • FastAPI
  • PostgreSQL
  • Clerk
  • Anthropic API

MarketEdge

Prediction-market divergence engine

In progress

An independent auditor comparing Kalshi prediction-market pricing against traditional sportsbook odds, detecting divergences and grading its own calibration over time.

  • Python
  • FastAPI
  • PostgreSQL
  • Prefect
  • httpx
  • Docker

ReturnTrack

Never miss a return window again

Shipped as a web app — app store release pending

Photograph a receipt and Claude's vision API extracts the items, then ReturnTrack tracks every return deadline and warranty countdown before it lapses.

  • Next.js
  • TypeScript
  • PostgreSQL
  • Prisma
  • Anthropic API
  • Clerk
All projects →