Skip to main content

​
What This Example Shows

  • A HierarchicalSwarm with an Investment Director coordinating three worker analysts: Comps, Macro, and Financing
  • How to feed a property address plus an ask price as the swarm task
  • How to extract the Director’s final memo as the deliverable
  • A real cost comparison against a broker’s analyst writing the same memo
This tutorial uses HierarchicalSwarm — the same primitive used by institutional research desks. To run memos on dozens of properties per day or wire this into a Slack bot for your acquisitions team, upgrade at https://swarms.world/platform/account.

​
Why This Matters

Real estate investment committees move slowly because the analyst floor moves slowly. Every deal — even a “let’s pass on this one” deal — eats two days of an associate’s time to produce a baseline memo: comps from the MLS or CoStar, a paragraph on the submarket, two financing scenarios, and a recommendation. Most of those memos end in a pass. The job is to push the trivial-no and trivial-yes deals through to a decision in minutes so the associate’s deep work stays focused on the maybes. A hierarchical swarm with three specialists and a Director does exactly that, and the output reads like an associate wrote it.

​
Step 1: Setup

pip install requests python-dotenv
export SWARMS_API_KEY="your-api-key-here"

import json
import os

import requests
from dotenv import load_dotenv

load_dotenv()

API_KEY = os.getenv("SWARMS_API_KEY")
BASE_URL = "https://api.swarms.world"

headers = {"x-api-key": API_KEY, "Content-Type": "application/json"}

​
Step 2: Define the Memo Team

The Director writes the final one-pager. The three workers each produce one section the Director will weave together. 
INVESTMENT_DIRECTOR_PROMPT = (
    "You are the Investment Director at a real estate private equity shop. "
    "You receive briefs from your Comps Analyst, Macro Analyst, and Financing Analyst. "
    "Write a single one-page investment memo in this exact structure:\n\n"
    "PROPERTY: <address>\n"
    "ASK: $<price>\n\n"
    "1. RECOMMENDATION: <PURSUE | PASS | PURSUE AT LOWER PRICE>\n"
    "2. THESIS: <three sentences on why>\n"
    "3. COMPS SUMMARY: <two sentences synthesizing the Comps brief>\n"
    "4. MARKET CONTEXT: <two sentences synthesizing the Macro brief>\n"
    "5. FINANCING: <two sentences synthesizing the Financing brief>\n"
    "6. KEY RISKS: <three bullets>\n"
    "7. NEXT STEPS: <one sentence — site visit, LOI, decline>\n\n"
    "Be decisive. The committee reads the recommendation line first."
)

COMPS_ANALYST_PROMPT = (
    "You are a Real Estate Comps Analyst. Given a property address and an ask price, "
    "produce a brief covering: three recent comparable transactions in the same submarket "
    "(estimated based on your knowledge of the area), price per square foot, cap rate range, "
    "and whether the ask is at, above, or below market. "
    "If you do not have specific transaction data, state the typical price-per-sf and cap-rate "
    "range for that property type and submarket. Keep it under 200 words."
)

MACRO_ANALYST_PROMPT = (
    "You are a Real Estate Macro Analyst. Given a property address, produce a brief covering: "
    "submarket fundamentals (population growth, employment, vacancy), the relevant "
    "interest-rate environment, supply pipeline, and the 18-month demand outlook for the "
    "property type. Keep it under 200 words."
)

FINANCING_ANALYST_PROMPT = (
    "You are a Real Estate Financing Analyst. Given a property and an ask price, produce a brief "
    "covering: two realistic capital-stack options (e.g. 65% LTV agency senior + sponsor equity "
    "vs. 75% LTV bridge + preferred equity), prevailing interest-rate ranges, estimated "
    "DSCR at stabilized NOI, and the equity check required for each. Keep it under 200 words."
)


def build_memo_swarm(address: str, ask_price: int, property_type: str) -> dict:
    return {
        "name": "Real Estate Investment Memo",
        "description": "Investment Director coordinating Comps, Macro, and Financing analysts.",
        "swarm_type": "HierarchicalSwarm",
        "max_loops": 1,
        "task": (
            f"Produce an institutional investment memo for the following deal:\n\n"
            f"PROPERTY ADDRESS: {address}\n"
            f"PROPERTY TYPE: {property_type}\n"
            f"ASK PRICE: ${ask_price:,}\n\n"
            "Each analyst writes their brief. The Investment Director then writes the final "
            "one-page memo with a clear recommendation."
        ),
        "agents": [
            {
                "agent_name": "Investment Director",
                "description": "Director — writes the final one-page memo.",
                "system_prompt": INVESTMENT_DIRECTOR_PROMPT,
                "model_name": "gpt-4.1",
                "role": "coordinator",
                "max_loops": 1,
                "max_tokens": 4096,
                "temperature": 0.3,
            },
            {
                "agent_name": "Comps Analyst",
                "description": "Comparable transactions and pricing.",
                "system_prompt": COMPS_ANALYST_PROMPT,
                "model_name": "gpt-4.1",
                "role": "worker",
                "max_loops": 1,
                "max_tokens": 2048,
                "temperature": 0.4,
            },
            {
                "agent_name": "Macro Analyst",
                "description": "Submarket fundamentals and demand outlook.",
                "system_prompt": MACRO_ANALYST_PROMPT,
                "model_name": "gpt-4.1",
                "role": "worker",
                "max_loops": 1,
                "max_tokens": 2048,
                "temperature": 0.4,
            },
            {
                "agent_name": "Financing Analyst",
                "description": "Capital stack, leverage, and DSCR.",
                "system_prompt": FINANCING_ANALYST_PROMPT,
                "model_name": "gpt-4.1",
                "role": "worker",
                "max_loops": 1,
                "max_tokens": 2048,
                "temperature": 0.4,
            },
        ],
    }

​
Step 3: Run the Memo for One Property

Paste in any deal that hits your inbox. Address, ask, property type — that is the entire input the swarm needs. 
def generate_memo(address: str, ask_price: int, property_type: str) -> dict:
    payload = build_memo_swarm(address, ask_price, property_type)
    response = requests.post(
        f"{BASE_URL}/v1/swarm/completions",
        headers=headers,
        json=payload,
        timeout=300,
    )
    response.raise_for_status()
    return response.json()


result = generate_memo(
    address="1200 Brickell Avenue, Miami, FL 33131",
    ask_price=48_500_000,
    property_type="Class A multifamily (212 units, built 2019)",
)

for output in result.get("output", []):
    print("=" * 60)
    print(output["role"])
    print("=" * 60)
    content = output["content"]
    if isinstance(content, list):
        content = " ".join(str(c) for c in content)
    print(str(content)[:800])

print(f"\nTotal cost: ${result['usage']['billing_info']['total_cost']:.4f}")
print(f"Execution time: {result['execution_time']:.1f}s")

​
Step 4: Extract Just the Final Memo

For your acquisitions pipeline you only want the Director’s one-pager — the analyst briefs are the audit trail. 
def extract_director_memo(result: dict) -> str:
    for output in result.get("output", []):
        role = output.get("role", "")
        if "Investment Director" in role or "Director" in role:
            content = output["content"]
            if isinstance(content, list):
                content = " ".join(str(c) for c in content)
            return str(content)
    return ""


memo = extract_director_memo(result)
print("\n--- COMMITTEE-READY MEMO ---\n")
print(memo)

# Persist it
with open("memo_1200_brickell.md", "w") as f:
    f.write(memo)
Wire this into a Slack slash command (/memo <address> <ask>) and your acquisitions team gets a committee-ready one-pager in under a minute. The analyst briefs sit in the full response if anyone needs to audit a specific section.

​
Real Cost vs. Analyst-Written Memo

ScenarioCost per memoTurnaroundCost per 100 deals screened
Investment Memo Swarm (4 agents, GPT-4.1)~$0.20~60 seconds~$20
Junior analyst writing the memo (fully loaded $120k)~$7001–2 days~$70,000
Outsourced underwriting shop~$300–$5003–5 days~$30,000–$50,000
You are not replacing the deal team — you are letting them screen 10x more deals before committing analyst hours, and giving the IC a structured starting point on every name that does make it through.

​
Next Steps