Harshit,Duggal

I build agent systems and AI products end to end.

Building Opal, an AI CRM for job hunting and matching — 28 users, 9,912 jobs surfaced, 29 resumes parsed, 1,258 matches.

Also building Dark Funnel, research infrastructure that turns one description into pipeline-ready prospect data.

01 / Introduction

I build AI systems from agent orchestration and infrastructure through the product people use. I own the path from ambiguous problem to production

02 / The work

Superflow

Free, open-source voice infrastructure for macOS that understands context and does work on your computer.

Superflow
  • Local speech recognition and local LLM inference (MLX, GGML, Metal) — capture, transcription, and reasoning on-device.
  • Context-aware behavior: current app, visible screen, selected text, files. Identical speech produces different output in Gmail, Slack, or a code editor.
  • Reasoning separated from execution: the model decides intent and structured actions; controlled integrations execute multi-step computer actions.
TaRuReTaSwAp
Visit site

Aether

An AI-native browser for frontier agents and humans.

  • A real browser frontier agents use, navigate, and work inside just like a human would: log in, research, operate SaaS tools, manage workflows, finish real work.
  • Each coding agent gets its own browser workspace tied to its worktree, with its own identity, sessions, and state.
  • Fleets of 50–100 agents in isolated workspaces, operating continuously across sites at machine scale.
  • Not another Computer Use wrapper. Computer Use gives agents a screen and mouse; Aether gives them the browser itself.
  • In development.
Sw
Visit site

Dark Funnel

Research infrastructure that turns one description into pipeline-ready prospect data in a single pass.

Dark Funnel
  • One description pulls people, companies, funding, hiring signals, and contact data together — no tool-hopping.
  • Funnel Agent replays your winning briefs on autopilot; every email verified honestly — missing beats invented.
  • Review the sheet, export clean CSV, launch Gmail sequences.
NeTyLaOpAgTaInoR
Visit site

Opal

AI CRM for job hunting that matches resumes to relevant jobs.

Opal
  • Real usage: 28 users, 9,912 jobs surfaced, 1,258 matches, 29 resumes parsed.
  • Resume-to-job matching with end-to-end ownership, from pipeline to interface.
NeTyTaLaAnLl
Visit site

Quantum

A tiny, beautiful browser with a sub-150 KB download, compared with Chrome at 1.2 GB.

  • Uses WebKit for rendering, storage, and HTTP caching without bundling a browser engine.
  • Native AppKit interface, with no Electron, npm, or third-party dependencies.
  • 149 KB DMG · 344 KiB app bundle · under 30 MB RAM.
Sw
Visit site

The stack I work with

The technologies I use across agent systems, infrastructure, data, and production AI.

Frontend
stack

  • TypeScript
  • Next.js
  • React
  • Tailwind CSS
  • GSAP
  • Motion
  • Vite

Backend
stack

  • Neon / Postgres
  • Drizzle
  • AWS
  • Google Cloud
  • Cloudflare
  • Bun
  • Docker
  • oRPC
  • Inngest
  • Sentry
  • Stripe

AI
stack

  • OpenAI Agents SDK
  • LangGraph
  • LangChain
  • Agent orchestration
  • LLM evaluation
  • Gemini Agents Kit

03 / How I think

Strong opinions, held loosely

  • I build for the product, not the resume

    Start with the problem. Keep the architecture as simple as the system allows. I work across orchestration, infrastructure, backend, and UI, then spend complexity where reliability or performance needs it.

  • I run a fleet of coding agents

    I use coding agents in parallel for implementation, debugging, and tests. I supervise the work, set architecture, review failures, and decide what ships. Agents increase throughput; ownership stays with me.

  • The harness matters

    I build around the model: tools, context, routing, memory, evals, recovery, and execution. The goal is reliable behavior in production, including when tools fail.

Care deeply about

Great products, reliable systems, production quality, and building things people love to use.

Care less about

Overengineering, unnecessary features, and complexity for its own sake.

04 / Contact

Work with me

I'm open to full-time engineering roles building agent systems and AI products, with ownership from architecture through production.