Manu Singhal

Posted on Feb 10

I Built a Human API Because My AI Agent Couldn't Walk Into a Building

#api #ai #python #webdev

The Problem

I was building an AI agent that helped people find PG accommodations in Bangalore. The agent could search listings, compare prices, make phone calls to landlords, and shortlist options. It was doing great — until a user asked:

"Can you actually go check if this PG is clean?"

And just like that, the agent was stuck.

AI can write code. It can make phone calls. It can send emails, summarize documents, and even book flights. But it can't physically walk into a building and check if the room smells like damp socks. It can't sit in a government office for 3 hours to get a property tax reassessment. It can't pick up documents from a lawyer's office across town.

AI hits a wall when the task requires a human body, a physical presence, or a real identity.

The Solution: ai2hum

So I built ai2hum — a Human API for AI agents. The concept is simple:

Your AI agent posts a bounty (task) via API
A KYC-verified human in India picks it up
The human physically completes the real-world task
Your agent gets back structured JSON with the result, proof photos, and notes

Think of it as the "UPI for Tasks" — one API call away from a verified human who can go where your model can't.

What Tasks Need a Human?

AI is incredible at digital tasks. But these need a real person:

Physical visits — Verify a PG, inspect a warehouse, survey a construction site, check product quality on the ground
Government portals — Navigate Aadhaar, passport, GST, IRCTC — portals that require real identity, CAPTCHAs, OTPs, and sometimes physically showing up
Document pickup & delivery — Collect signed contracts, drop off paperwork, courier things across town
On-ground verification — Is this shop actually open? Does this address exist? Is the product real?
Expert reviews — Legal analysis, compliance checks, quality assessment that needs human judgment
Meetings & events — Attend on behalf, take notes, represent at a government hearing

All across 12+ Indian cities, with payments via UPI.

Integration: 3 Ways, Under 5 Minutes

Option 1: MCP Tool (for Claude, GPT, and LLM agents)

Drop ai2hum as a tool for any MCP-compatible AI agent. Zero code changes:

{
  "mcpServers": {
    "ai2hum": {
      "command": "npx",
      "args": ["@ai2hum/mcp-server"],
      "env": {
        "AI2HUM_API_KEY": "sk_live_your_key_here"
      }
    }
  }
}

Your agent now has 5 tools: create_task, get_task, list_tasks, list_workers, and cancel_task.

Option 2: REST API

const response = await fetch('https://api.ai2hum.ai/v1/tasks', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer sk_live_your_key_here',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    type: 'physical_visit',
    title: 'Verify PG accommodation in Koramangala',
    description: 'Check room size, WiFi speed, food quality',
    location: 'Bangalore, Karnataka',
    reward: 500,
    urgency: 'high',
    callback_url: 'https://your-app.com/webhooks/ai2hum'
  })
});

const task = await response.json();
// { id: "BNT-017", status: "ASSIGNED", eta: "30-60 min" }

Option 3: Python SDK

from ai2hum import Ai2hum

client = Ai2hum(api_key="sk_live_your_key_here")

task = client.tasks.create(
    type="phone_call",
    title="Call BBMP about property tax dispute",
    description="Inquire about reassessment for ID KA-BLR-2024-88431",
    location="Bangalore, Karnataka",
    reward=300,
    min_trust_score=80
)

# task.status = "ASSIGNED"
# task.matched_worker.name = "Priya S."
# task.match_score.total = 95

How the Matching Works

When you post a task, ai2hum's matching engine scores available humans on:

Skills — Does the human have the right capabilities?
Location — How close are they to the task?
Trust score — Track record from previous tasks
Availability — Are they free right now?

You can even re-hire a specific worker with preferred_worker_id if they did great last time.

The Trust Layer

Every human on ai2hum is:

KYC-verified with Aadhaar (India's biometric ID)
Trust-scored after every completed task
Paid instantly via UPI — no waiting for bank transfers
Escrow-protected — payment is held until the task is verified complete

No credential sharing. No impersonation. Full audit logs for every task.

What You Get Back

When a human completes a task, your agent receives structured JSON:

{
  "id": "BNT-017",
  "status": "COMPLETED",
  "type": "physical_visit",
  "result": {
    "summary": "PG verified. Room is 12x10, WiFi 45mbps, meals included.",
    "checklist": {
      "room_clean": true,
      "wifi_working": true,
      "food_quality": "good",
      "safety": "gated entry with CCTV"
    }
  },
  "proof": {
    "photos": ["https://..."],
    "notes": "Spoke with owner. 3-month minimum stay required.",
    "completed_at": "2026-02-10T14:30:00Z"
  },
  "worker": {
    "id": "WRK-001",
    "trust_score": 96
  }
}

Structured. Machine-readable. Ready for your agent to act on.

Pricing

Plan	Price	Tasks/Month
Free	Rs 0	5 tasks
Starter	Rs 2,999/mo	50 tasks
Growth	Rs 9,999/mo	200 tasks
Enterprise	Custom	Unlimited

Each task also has a bounty reward (₹100–₹5,000) paid to the human.

Why I Built This

Every AI builder eventually hits the same wall: the physical world. Your agent can find a PG listing but can't visit one. It can generate a GST filing but can't sit in the government office when they ask you to "come in person." It can plan a logistics route but can't verify if the warehouse actually exists.

The gap isn't intelligence — AI has plenty of that. The gap is presence. A body that can walk into a room, eyes that can inspect a product, hands that can collect a document.

ai2hum is the bridge. AI does 80% of everything — we handle the 20% that needs a human on the ground.

DEV Community