jp.
Resume (opens in a new tab)
JASH PARMAR/INDIA · REMOTE

AI Evaluation Engineer
& Full-Stack Developer.

I build software-engineering benchmarks for AI coding agents, reliable LLM evaluation systems, and full-stack products.

View selected work Resume (opens in a new tab)
CURRENTLY

Advanced AI Engineer at Crossing Hurdles

LLM Trainer at Handshake AI

01

Selected work

Practical systems. Considered engineering.
workflow-builder↗
Workflow Builder with a six-node palette, Text and Upload Image nodes on a canvas, and a run history panel.
Canvas → connections → executionReact Flow / Zustand
01AI tooling / Visual editor

Workflow Builder

Compose multi-step LLM and media workflows on a visual, node-based canvas.

  • Six node types with typed input/output handles and data propagation between connected nodes.
  • Individual-node and full-workflow runs, execution states, and run history managed in Zustand.
  • Configurable Gemini models and prompts; test coverage with Vitest.

React · TypeScript · React Flow · Zustand · Tailwind CSS · shadcn/ui · Vite · Vitest

Prototype: node execution currently returns simulated outputs.

Live demo (opens in a new tab)GitHub (opens in a new tab)
02LLM systems / Retrieval

Smart Screen Recorder

Turn historical screen activity into a searchable semantic knowledge base.

  • Change-aware screen capture, OCR extraction, and semantic embeddings stored locally in ChromaDB.
  • Gemini rewrites questions; vector retrieval supplies screen context for grounded RAG responses.

Python · OpenCV · Tesseract OCR · Sentence Transformers · ChromaDB · Gemini API

GitHub (opens in a new tab)
FROM PROBLEM
TO PATTERN.
Sliding WindowBinary SearchDynamic ProgrammingTwo Pointers
Local detection + AI guidance
PatternSense mobile app with a DSA problem statement field and Detect Pattern action.
03AI-assisted / Mobile

PatternSense

Find the algorithmic pattern behind a coding problem, then understand the approach.

  • Local rules rank patterns by confidence, with a token-overlap matcher for secondary hints.
  • An Express backend uses Gemini JSON mode for examples, approaches, and time/space complexity.

React Native · Expo · Expo Router · TypeScript · Node.js · Express · Gemini API

GitHub (opens in a new tab)
02

Experience

Real tasks. Reproducible results.
Jul 2026 — PresentRemoteCurrent

Advanced AI Engineer

Crossing Hurdles

Project Sentinel Ultra · Terminus 3

  • Develop software-engineering benchmarks for frontier AI coding agents, with an emphasis on correctness and reproducibility.
  • Design Harbor-format tasks with oracle implementations, deterministic verifiers, Docker environments, and automated evaluation pipelines.
Benchmark engineering & evaluation infrastructure
Jul 2026 — PresentRemoteCurrent

LLM Trainer

Handshake AI

Project Dynamo · Terminal-Bench 2

  • Author end-to-end evaluation tasks for Terminal-Bench 2, spanning debugging, infrastructure, systems programming, repository workflows, and data processing.
  • Build oracle solutions, Docker environments, deterministic verification pipelines, and automated test suites.
Terminal-Bench task authoring & testing
Feb 2026 — Jun 2026Remote

Software Development Engineer Intern

Xelron AI

Terminus · Finch · Project 1307

  • Contributed to prompt engineering and benchmark evaluation for Terminus and Finch.
  • Evaluated Gemini responses on Project 1307 with Uber AI Solutions, collaborating with distributed teams on enterprise AI evaluation workflows.
AI evaluation & model responses
May 2025 — Jul 2025Remote

Full-Stack Developer Intern

Flitzen Technologies UK Limited

Golf competition platform

  • Led development of a responsive golf competition platform with subscriptions, QR integration, and Winners’ Circle features.
  • Built community-focused product functionality and responsive interfaces.
Full-stack product development
03

Engineering focus

AI evaluation

Making agent capabilities measurable.

Software-engineering benchmarks, coding-agent evaluation, deterministic verifiers, oracle implementations, automated evaluation pipelines, prompt evaluation.

LLM systems

Connecting models to useful context.

Retrieval-augmented generation, semantic search, embeddings, vector databases, Gemini API, prompt engineering.

Software engineering

Building the systems around the model.

React, TypeScript, Node.js, Express, Python, Docker, PostgreSQL, Linux, and testing.

04

More projects

A few more things I’ve built.
Earlier experiments
05

A little about me

I’m Jash, a Computer Science engineer from IIIT Kota, working at the intersection of AI evaluation and software engineering.

My recent work centers on software-engineering benchmarks for AI coding agents, deterministic verification, and reproducible evaluation environments. I also enjoy building full-stack products and practical LLM applications.

I’m interested in remote opportunities with technically difficult, measurable problems.

Tools I work with

Languages

Python · TypeScript · JavaScript · C++ · Java

Frontend

React · Next.js · React Native · Tailwind CSS · React Flow · Zustand

Backend & data

Node.js · Express · Django · REST APIs · PostgreSQL · MySQL

AI / LLM

LLM evaluation · AI benchmarking · RAG · Sentence Transformers · Semantic search · ChromaDB · Gemini API

Engineering & tools

Docker · Linux · Git · GitHub · Vitest · Postman · Vercel · Render

LET’S BUILD SOMETHING USEFUL

Let’s work together.

I’m open to remote AI / LLM evaluation, software engineering, and full-stack opportunities.

[email protected]