lord of the land

planing
first go
codex errors handaling
codex token run out
added judges on git repo
missed the token option
report first page
property for sale open to bids
same property we have listed "off market" on other web sites
us gov tenderes on map
tender ai analysis

Inspiration

For the last year Im vibe-coding my dream app (www.lordoftheland.com) a single place to bring planning and real-estate data together. It started in Israel, then followed me to Greece, and later to the US and Spain, growing city by city as I added places I liked that had real-estate potential.

The problem I kept hitting is fragmentation. A buyer, investor, appraiser, broker, attorney, or developer often has to search across many government and city websites just to understand one property. Federal listings, parcel data, zoning rules, market trends, maps, and planning documents live in separate systems, in different formats, and each demands professional knowledge to interpret.

Lord of the Land was inspired by a simple question: What if you could get it all in one place, as one package? That was the personal project.

When the Kaggle / Google AI course came along last month, I wanted a sharper problem to build an agent around — and asked what value I could add on top of the data. What a real-estate investor really wants is the best-value deals and the properties still under the radar, out of the wide public eye. That question is where the SAM.gov idea came up: a quiet corner of government tenders — the government selling property, or seeking space to lease — that's fully public yet hard to read at human speed. So I built on it as a Google ADK agent that gathers SAM data from different sources, so anyone hunting a federal real-estate opportunity can reach the relevant facts, risks, and opportunities from one intelligent interface.

For the OPEN AI BUILD WEEK I took the base app I already had and used Codex with GPT-5.6 to build a more powerful tool on top of it: three GPT-5.6 agents that process the data, read the documents, and answer the user's questions - combining professional real-estate knowledge with AI, public data, mapping, and automated analysis, layered on the existing Google ADK backbone.

Its purpose isn't to replace professional judgment — it's to make the first stage of tender due diligence faster, clearer, and more accessible and even fun.

What It Does

Lord of the Land is an AI-powered federal real-estate intelligence platform. The application brings together several types of public information and cross check them, such as:

Federal real-estate opportunities
Government property sales and leasing requirements
Parcel and location data
Zoning and planning information
Housing-market indicators
Interactive map layers
GEE (google earth engine) data layers for real estate data analysis
AI-assisted search and analysis

Users can explore opportunities on a map, review available property information, and ask questions about locations, market conditions, zoning, and government real-estate activity.

For OpenAI Build Week, we are extending the existing platform with a new feature:

GPT-5.6 Federal Opportunity Analyst

The new workflow allows a user to select a federal real-estate opportunity and request a structured due-diligence analysis.

The analyst is designed to:

Summarize the opportunity.
Review the available property and location information.
Add zoning and market context where data is available - 4 cities are wired.
Identify risks, missing information, and important assumptions.
Recommend practical next steps.
Produce a clear Pursue, Review, or Skip assessment.
Show the sources used and distinguish source facts from AI interpretation.

When information is unavailable, the system is instructed to state that clearly rather than inventing an answer.

How We (me and my AI coding agent) Built It

In thw Google & Kaggle AI Agents Capstone project —i started the Google ADK multi-agent system with an LLM layer on top (gpt 5 mini). That gave me the initial architecture, public-data integrations, map interface, and deployed application.

During this OPEN AI BUILD WEEK, the existing application got meaningfully extended using Codex and GPT-5.6. Codex is being used to:

Inspect and understand the existing codebase
Design the new opportunity-analysis workflow
Connect GPT-5.6 to existing property and market data tools
Build structured model outputs
Create backend endpoints and user-interface components
Add validation, error handling, and tests
Document the architecture and deployment process
Separate pre-existing work from the new Build Week implementation

The project uses a deliberate human-and-agent collaboration model. Codex accelerates implementation, debugging, testing, and documentation. Human expertise defines the professional logic: which facts matter, how uncertainty should be presented, what constitutes a meaningful real-estate risk, and how recommendations should be communicated responsibly. and finish up with cloud

Challenges

The most time killing and annoying was building the interface through an AI coding agent was harder than it sounds. The agent writes the map layers, the pins, and the buttons — but it can't see the result, and I often struggled to put into words what felt wrong. Getting pins to show correctly on the map, naming the buttons so the action was obvious (Quick check vs Full deal review), making the 3D↔2D toggle behave, and giving honest progress feedback each took a loop of render → look → describe → fix, which was sometimes genuinely exhausting. But along the way I started to understand the code a little myself — both the Python and the index.html — enough to make small changes on my own: text sizes, element colors, spacing, nudging things left and right, and starting to recognize functions and what they do. That turned the process from pure back-and-forth into something I could begin to steer directly.

Another main challenges was data fragmentation. Different government and municipal sources use different identifiers, geographic structures, schemas, update cycles, and access methods. Some locations have detailed parcel and zoning information, while others provide only limited data. downloading for each supported city the zooning files and proces them, embading, chunks and load to chroma vector stor for RAG pipeline to serve both user and agents.

A second challenge was avoiding false certainty. Real-estate analysis often involves incomplete records, outdated documents, jurisdiction-specific rules, and facts that require professional verification. The product therefore needs to distinguish between:

Verified source data
AI interpretation
Professional assumptions
Missing information
Issues requiring further investigation

Another challenge was extending an existing product within a short hackathon period. Instead of rebuilding the entire platform, i focused on one complete and useful workflow that demonstrates how GPT-5.6 and Codex can transform disconnected public data into an actionable decision-support experience.

What have i Learned

I learned that the greatest value of AI in real estate is not simply generating text.

The real value comes from combining:

Reliable data retrieval
Tool use
Structured reasoning
Clear source attribution
Domain expertise
Honest handling of uncertainty
gis data, rest api, post gis - finding sources and wire them

We also learned that an effective coding agent is most powerful when it works within clear product boundaries. Codex accelerated the engineering process, but the quality of the result depended on precise requirements, continuous testing, and professional decisions about the user experience.

Most importantly, i learned that complex real-estate information can become significantly more understandable without oversimplifying it.

What’s Next

The next stage is to expand the analyst into a broader land due-diligence system that can support additional jurisdictions, planning documents, valuation indicators, environmental constraints, infrastructure information, and transaction data. Implement GSA Public Buildings Service Pricing Desk Guide as a pricing-methodology grounding source.

Our long-term vision is to create an intelligent layer for land and real-estate decisions worldwide: