BURAN Digital Economic Office

Inspiration

One day I saw how much routine work lawyers have to deal with. A large amount of their working time can be spent processing legal documents. Before a lawyer can prepare a case and make a professional decision, the documents need to be read, organized, compared, checked for contradictions, and connected to the facts and chronology of the case.

This gave me the idea to create BURAN, a digital platform where specialized digital professionals perform structured professional work with clearly defined responsibilities and controls.

My goal is to build a system where digital professionals can perform up to 95% of structured production work, while people keep professional control and responsibility where they are needed.

BURAN was already under development before this hackathon was announced. I am not presenting the whole platform as something created during the competition.

For this hackathon, I am developing one of the important missing parts of BURAN: the Digital Economic Office. It is intended to become an integral part of the existing BURAN platform.

I saw an opportunity to develop this new part using Gemini and Google Cloud.

What it does

The Digital Economic Office is not a general-purpose AI assistant.

It is being designed as a governed group of specialized digital professionals. Each professional has a specific responsibility and must produce a defined result.

The first problem we are solving is understanding the real operating costs of a digital company.

BURAN uses servers, storage, backups, OCR, AI providers, payment services, security services and other infrastructure. Some costs are fixed. Some depend on actual usage. Some change as the platform grows. Other costs cannot be accepted as confirmed until we have a real quotation or information from a reliable official source.

The Digital Economic Office is intended to collect and organize this economic information and distinguish between:

• verified information;
• information that still needs verification;
• Owner Decisions;
• costs that require a quotation;
• information that is still unknown.

An unknown cost must not automatically become zero, and an AI-generated estimate must not automatically become an accepted financial fact.

This is also why we do not treat every digital professional as an LLM.

Some work can be performed by deterministic software, rules or local tools. Gemini is used where research, reasoning or interpretation is genuinely useful.

How we built it

The main architectural principle is separation of responsibilities.

Instead of giving one AI model a large instruction to research, analyze, calculate, verify and approve everything, we divide the work between specialized digital professionals and deterministic components.

For example, the workflow can include:

Research specialist

Searches for missing economic information and collects evidence from external sources. Gemini can help with research and interpretation.

Verification specialist

Checks the source, conditions and supporting evidence. Information found by the research specialist does not automatically become an accepted financial fact.

Economic specialist

Receives verified information and performs a defined professional economic task.

Deterministic BURAN components

Perform exact calculations, apply defined rules and check formal conditions. We do not need a generative AI model to perform ordinary arithmetic.

Control and certification

Check that the required evidence is present and that the defined process has been followed before a result can be accepted.

For the hackathon, Google Cloud provides the environment for the new Gemini-based capabilities and the services they need.

We are not moving the whole BURAN platform to Google Cloud and we are not creating a second version of BURAN.

The new Google Cloud capability is being developed as part of the Digital Economic Office and integrated into the existing BURAN architecture.

The principle is simple: Gemini performs the intellectual work where generative AI is useful. Deterministic software performs the work where AI is not needed. BURAN controls the overall workflow and the rules for accepting results.

Challenges we ran into

One of the biggest challenges was deciding what an AI system should be allowed to do.

It is easy to connect an LLM to an application and ask it to produce an answer. That is not enough for a professional digital company.

If a digital specialist finds a price on the internet, should that price immediately become an official cost in BURAN? We decided that it should not.

Research and authority need to be separated.

Another challenge was understanding the real cost structure of a digital company. A digital professional is not the same thing as an LLM call. Some digital professionals may use Gemini, while others can work with deterministic code, rules or local tools.

We also had to distinguish between fixed costs, usage-based costs, costs that change with scale, and costs that are still unknown.

This led to an important rule: missing economic information must remain visible as missing information. We should not invent a convenient number simply to complete a calculation.

Accomplishments that we're proud of

The biggest accomplishment for me is that BURAN has grown beyond the idea of using AI to answer legal questions.

We are building it as a digital company with specialized digital professionals, defined responsibilities, production controls and a real workflow.

Before starting the Digital Economic Office, we had already developed a substantial part of the BURAN legal production system.

As part of our current end-to-end work, a controlled synthetic case containing 14 documents and 24 pages successfully passed the document readability stage. All 24 pages were recovered using local OCR without an external AI provider call, and the case reached the point where it was ready for pricing.

That result was important because it also confirmed one of our architectural principles: we should not use an AI provider simply because it is available. If a less expensive deterministic or local tool can correctly perform the task, BURAN should use it.

For this hackathon, I am proud that we are not building a separate demonstration application. We are using the competition to develop a real missing part of BURAN that can remain in the platform after the hackathon.

What we learned

One of the most important things we learned is that a digital professional and an AI model are not the same thing.

The professional role defines the responsibility. Gemini is a tool that can be used by that role when its capabilities are needed.

We also learned that different types of work need different controls.

Research is different from verification. Verification is different from calculation. Calculation is different from certification.

Combining all of these responsibilities into one unrestricted AI agent may be simple to demonstrate, but it is not the architecture we want for BURAN.

We also learned that deterministic software is very important in an AI system. If something can be calculated exactly, it should be calculated exactly.

Finally, we learned that uncertainty must remain visible. If a cost is unknown, the correct answer can be "unknown". It is better to identify missing evidence than to create a confident but unsupported financial number.

What's next for BURAN Digital Economic Office

The next step is to complete the real operating cost structure of BURAN.

We are identifying:

• variable costs connected to actual case processing;
• fixed monthly operating costs;
• annual and one-time costs;
• costs that appear as the platform scales;
• tax and statutory items that require separate verification;
• costs for which a real quotation or official source is still required.

After this foundation is complete, the Digital Economic Office can move to the next economic stages: full cost analysis, client pricing, profitability and break-even analysis.

For the hackathon, the immediate goal is narrower.

We want to implement and demonstrate a real workflow in which specialized digital professionals use Gemini and Google Cloud for a defined economic task, collect supporting evidence, pass the result through verification and control, and return a structured result into the BURAN workflow.

The goal is not to show that Gemini can produce another answer.

The goal is to show how Gemini can become a useful professional capability inside a governed digital company where responsibilities are separated, evidence can be checked, and important economic decisions remain explainable and auditable.

Built With

• cloudrun
• gemini
• googkecloud
• googleadk
• googleai
• vertexai