# neelabhbuilds.com

> Machine-readable copy of this site, generated from the same source as the
> human view. If the two ever disagree, this file is the bug.

## who

Neelabh Kumar. San Francisco.

Builds AI agents, from model to infrastructure.

## now

### Training a small model to run a house
date: 2026-09-27
url: https://www.neelabhbuilds.com/writing/training-a-small-model-to-run-a-house

I trained Qwen3-1.7B, an open model small enough to run on a consumer GPU, to control a house through Home Assistant. On the Home Assistant community's voice-command test it went from 75.0% to 89.8%.

## work

### hua-1.7b
year: 2026
link: https://www.neelabhbuilds.com/writing/training-a-small-model-to-run-a-house

A 1.7B model trained to run a house through Home Assistant.

### Computer use
year: 2026
link: https://www.neelabhbuilds.com/projects/computer-use

A local 8B computer-use model on 50 OSWorld tasks.

### Postern
year: 2026
link: https://getpostern.com

One gateway between your agents and your data, on hardware you own.

### ClawLens
year: 2026
link: https://www.neelabhbuilds.com/projects/clawlens

Observability and guardrails for OpenClaw agents.

### JitAPI
year: 2026
link: https://www.neelabhbuilds.com/projects/jitapi

An MCP server that finds only the endpoints a task needs.

### PunkClanker
year: 2026
link: https://www.neelabhbuilds.com/projects/punkclanker

An agent fleet that ran WhatToStream on its own.

### Apple Health Bot
year: 2024
link: https://www.neelabhbuilds.com/projects/apple-health-bot

Ask your Apple Health data questions in plain English.

### Morsand
year: 2026
link: https://www.neelabhbuilds.com/projects/morsand

A 149-page manuscript about my ancestral village, digitized.

## writing

- Training a small model to run a house · 2026-09-27 · https://www.neelabhbuilds.com/writing/training-a-small-model-to-run-a-house

## upstream

- 2026-09-23 · allenporter/home-assistant-datasets #343 · open
  A row for hua-1.7b on the leaderboard, on all three test sets.
  https://github.com/allenporter/home-assistant-datasets/pull/343
- 2026-09-18 · xlang-ai/OSWorld #593 · open
  Four task checkers passed without the task being done. They now check the result.
  https://github.com/xlang-ai/OSWorld/pull/593
- 2026-09-14 · xlang-ai/OSWorld #582 · merged
  The harness dropped code from replies cut off before the closing fence. Now it keeps it.
  https://github.com/xlang-ai/OSWorld/pull/582
- 2026-09-12 · xlang-ai/OSWorld #592 · open
  A failed setup step was logged as complete and the task ran anyway. Now setup failures fail the task.
  https://github.com/xlang-ai/OSWorld/pull/592
- 2026-08-27 · xlang-ai/OSWorld #581 · open
  The task detail page crashed on timestamps with microseconds. Fixed.
  https://github.com/xlang-ai/OSWorld/pull/581
- 2026-04-12 · openclaw/openclaw #64979 · merged
  Approval callbacks in the Telegram plugin shared a lane with the agent turn that was waiting on them. Gave the callbacks their own lane.
  https://github.com/openclaw/openclaw/pull/64979

## elsewhere

github: https://github.com/nk3750
x: https://x.com/neelabhbuilds
linkedin: https://www.linkedin.com/in/neelabh-kumar/
email: mailto:neelabhkumar@outlook.com