akashnaren/agent-ui-human

Agent UI Human

Human-written interface preferences for agents. Companion to akashnaren/agent-ui-sft (synthetic multi-turn tool traces). This set is the opposite shape: one person-shaped request per row, a preferred UI, and a short rationale.

Author	Akash Premkumar (akashnaren)
License	Apache-2.0
Files	train.jsonl (50), test.jsonl (10), agent_ui_human.csv (60 full)
Mirror	Kaggle: akashpnaren/agent-ui-human

How it was made

Public question: akashnaren/agent-ui-metrics — what is the most efficient UI for agents to talk to tools?

Sixty original prompts were written one at a time for a lab setting (local git, named metrics, on-screen tabs, Hub/Kaggle-style wizards). They are not a template expanded with paraphrases, not scraped tickets, not Tesla service logs, and not customer data.

Labeling rule of thumb used while writing:

1. Bytes already on disk → cli
2. Stable named resource (device, run, metric) → structured_api
3. Answer already on an open page (banner, tab, widget) → dom_click
4. Product only exposes a wizard (title, license, visibility) → form

No inter-annotator study. No production capture. Small preference file so HF/Kaggle profiles have a human artifact next to the synthetic SFT set.

Schema

field	type	meaning
id	string	h-001 … h-060
preferred_ui	string enum	cli | structured_api | dom_click | form
domain	string	local_ops, telemetry, diagnostics, workflows, web_ui, authoring
prompt	string	Person-shaped request
rationale	string	Why that interface is the cheap one for this request

Split rule

Ids whose number is divisible by 6 go to test (h-006, h-012, …) → 10 test / 50 train. CSV holds all 60 rows.

Label mix (published train+test)

Checkable by loading the CSV / JSONL. Train is roughly balanced across four UIs; test is small (10 rows) and not stratified to equal counts.

How to load

from datasets import load_dataset

ds = load_dataset("akashnaren/agent-ui-human")
print(ds["train"][0])

CSV:

import pandas as pd
df = pd.read_csv("agent_ui_human.csv")
print(df["preferred_ui"].value_counts())
print(df["domain"].value_counts())

Classifier-shaped columns:

texts = ds["train"]["prompt"]
labels = ds["train"]["preferred_ui"]

Example rows (from published train)

id	preferred_ui	domain	prompt (short)
h-001	cli	local_ops	Show which python files changed since last commit.
h-011	structured_api	telemetry	Pack voltage for device lab-104 over last 15 minutes.
h-021	dom_click	web_ui	Open the Metrics tab on the already-loaded lab dashboard.
h-031	form	authoring	Create a public dataset titled Agent UI Human, Apache-2.0.

What it is not

• Not vehicle telemetry from the field.
• Not a dump of anyone's inbox.
• Not LLM-paraphrased clones of one seed (contrast agent-ui-sft, which is synthetic traces).
• Not Kimi / Moonshot weights or a proprietary fine-tune dump.
• Not a benchmark that generalizes beyond this 60-row lab preference file.

Intended use

Train or evaluate a small UI-mode classifier; read rationales for notes on interface cost. Pair with ui-mode-router as a related (synthetic-trained) baseline — that model was trained on agent-ui-sft, not on this human set.

Limitations

• 60 rows; single author; no IAA.
• English only.
• Domains are lab-shaped (local_ops, authoring, …), not production traffic.
• Test split is tiny (10 rows).

Links

• Synthetic companion: https://huggingface.co/datasets/akashnaren/agent-ui-sft
• Pairwise prefs (same labeling rules): https://huggingface.co/datasets/akashnaren/agent-ui-mode-pairs
• Flat efficiency bakeoff table: https://huggingface.co/datasets/akashnaren/agent-ui-efficiency-scores
• Question repo: https://github.com/akashnaren/agent-ui-metrics
• Collection: https://huggingface.co/collections/akashnaren/agent-ui-lab-6a9a8e06fec692165b0b3c07
• Kaggle mirror: https://www.kaggle.com/datasets/akashpnaren/agent-ui-human
• Model (trained on SFT, not this set): https://huggingface.co/akashnaren/ui-mode-router
• Personal site: https://akashnaren.github.io/
• ORCID: https://orcid.org/0009-0001-8877-9527
• Cursor: https://cursor.com/@akashpn
• Fleet / bot page: https://akashnaren.github.io/bot/