Skip to content

Ashish Khare — I'm Ashish Khare, a software engineer building products, systems and AI agents. I work across backend services, full-stack products and the infrastructure that keeps AI systems running — currently at Quinn.

Portrait of Ashish Khare

Education

Kamla Nehru Institute of Technology, Sultanpur

B.Tech, Computer Science and Engineering

2021 — 2025

CGPA 8.6 · Division I (Hons)

Achievements

Contact

Email

ashish.khare3110+portfolio@gmail.com

Elsewhere

Selected Work

01

Seline

AI Harness Platform

A Go platform that runs a personal AI agent for every user inside its own container — booted on signup, kept alive between tasks, and driven over Telegram.

  • One isolated container per user, spawned on signup and lifecycle-managed from there
  • Always-on execution with configurable heartbeats, scheduled jobs and mid-task interrupts
  • Sub-agents, an entity graph and a compacting memory layer behind a tool-calling loop
  • Go
  • Gin
  • Docker
  • PostgreSQL
  • TypeScript
  • Next.js
Read the case studymeetseline.com

02

Radius

Bookmark Search & MCP Server

A self-hosted Python MCP server that turns a flat pile of saved bookmarks into a relational corpus an AI agent can actually query.

  • FTS5 lexical search and RapidFuzz similarity over real tables, not a JSON blob
  • Streamable-HTTP MCP server built on FastMCP, guarded by RS256 JWTs and per-tool scopes
  • Runs locally or as a single Python function on Vercel against one Turso database
  • Python
  • FastMCP
  • Turso / libSQL
  • SQLite FTS5
  • RapidFuzz
  • JWT
  • Vercel
Read the case studyhalf-the-diameter.vercel.appGitHub

03

Playgroundish

Coding Practice Platform

A full-stack practice platform built around messy, real-world engineering problems — an in-browser editor and submission flow on one side, an authoring portal on the other.

  • Interactive in-browser code editor with a problem-to-submission flow
  • Admin portal for authoring and managing the problem catalogue
  • Built around real-system problems rather than interview brain teasers
  • TypeScript
  • Next.js
  • React
  • PostgreSQL
Read the case studyplaygroundish.vercel.app

Experience

  1. Quinn

    Sep 2025 — Present

    Software Engineer — Remote · Bengaluru

    • Designed a closed-loop QC pipeline that folds reviewer feedback back into its own prompts, cutting manual tuning and moving QA workflows 3× faster.
    • Built an image-to-metadata OCR pipeline on Gemini and BullMQ that replaced a manual extraction workflow at 96% accuracy.
    • Wrote a spatially-aware background engine that segments scenes into body-aligned regions, reducing visual artifacts by 60%.
    • Architected a pose-aware judgement pipeline that buckets expressions and conditions generation on anchor images for consistent output.
    • Scaled the Shopify plugin platform with reusable templates, lifting adoption 25% and onboarding 10+ customers.

  2. Stoira

    Jul 2024 — Sep 2024

    Software Engineer Intern — Remote · Bengaluru

    • Rebuilt list rendering on FlashList to hold 60fps through long, media-heavy feeds.
    • Engineered an audio system with queue management, background playback and lockscreen controls.
    • Shipped multi-provider authentication with deep linking, backed by MongoDB user profiles.
    • Built a cross-platform UI system of 20+ reusable components in Unistyles.

  3. Frontend Hire

    May 2024 — Jun 2024

    Design Engineer Intern — Remote · Hyderabad

    • Built the React Query and Zustand data layer for a React Native app, plus the reusable component set on top of it.
    • Shipped a markdown publishing flow with a custom Remark plugin, IntersectionObserver-driven navigation and Satori-rendered banners.

Stack

Languages
TypeScript, Go, Python, SQL
Backend & Systems
Node.js, Gin, FastAPI, FastMCP, REST APIs, Distributed Systems, Docker, Async Processing
Databases
PostgreSQL, MongoDB, Turso
Frontend
React, Next.js, React Native, React Query, Zustand
Tools & Platforms
Git, BullMQ, Supabase, GCP, Postman, Vercel

Writing

An engineering notebook — papers implemented to understand them, systems taken apart, and notes from things that broke in production.

Experiments

Smaller builds and older work — libraries, apps and model training runs kept around because the ideas in them still hold up.

  • Voodoo

    AI search & agent platform

    An AI search and agent platform built on diffusion LLMs, with parallel generation, multi-source search and agentic workflows.

    • TypeScript
    • Next.js
    • AI SDK
    • Convex

  • Wingman

    On-device VLMs for Android

    A mobile app that runs vision-language models locally on-device, backed by llama.cpp and a FastAPI service.

    • React Native
    • FastAPI
    • llama.cpp

  • Altar

    Microblogging app

    A microblogging app with real-time updates, user profiles and bookmarks, including a designed invite system.

    • TypeScript
    • Next.js
    • Convex

  • Commit

    Daily task tracker

    A daily task-tracking app built from its own Figma case study — design system, auth and animation included.

    • React Native
    • React Query
    • Expo

  • folder-routing

    Express routing library

    A small Express wrapper that maps a directory tree to routes, Next.js style, with async middleware support.

    • JavaScript
    • Express

  • Tetrapack

    Notion-style block parser

    A JavaScript parser that renders Notion-style blocks in isolation, bundled with esbuild and documented in Storybook.

    • JavaScript
    • esbuild
    • Storybook

  • Flower GAN

    Deep learning notebook

    A DCGAN trained from the 2015 Radford et al. paper to generate 64×64 flower images.

    • Python
    • PyTorch