OpenAI 2026 hackathon

TaxGraph for Software & AI Services

Turn a cross-border AI deal into a source-backed map of tax touchpoints, missing facts, and questions for professional review — before the first invoice is issued.

Solo project by Kindly-Peacefull Stepanov · 1 likes · 0 comments

Archive position — measured, not model output

1 like on Devpost

506 of the 7,856 archived projects have more likes, and 1,758 share exactly 1 — so this project's #2,042 place in the like-ranked listing is a tie-break inside that group, not a ranking.

Projects (log scale)

1
10
100
1k
10k
05,592
11,758
2285
3–4132
5–975
10+14

Likes on Devpost. ▲ marks this project's group.

Show the figures
LikesProjectsShare of archive
05,59271.2%
11,75822.4%
22853.6%
3–41321.7%
5–9751.0%
10+140.2%
Devpost like counts for all 7,856 archived projects, captured when this archive was built.

Executive Summary

What the company appears to be

TaxGraph for Software & AI Services is a self-reported tool that uses GPT-5.6 and structured rules to analyze cross-border software or AI transactions before invoicing. It aims to help small companies understand tax implications, missing facts, and questions for professional review — without generating a final tax opinion.

What changed

The author reports building this as part of an OpenAI hackathon project, using Codex for development over three days. The tool is described as deployed in production with GPT-5.6 running in production and VAT ID checks via VIES.

Single most important open question

Is there any evidence of actual usage or traction from real users beyond the author’s own testing?

Back to contents

What The Product Actually Is

The description states that TaxGraph:

  • Analyzes cross-border software or AI transactions before the first invoice is issued.
  • Uses GPT-5.6 to extract and structure facts from user input (form, free text, contract excerpt).
  • Separates a mixed product into components such as SaaS access, integration, hosting, support, training, and licensing.
  • Tracks where each fact came from, identifying conflicts between inputs.
  • Applies twelve fixed tax rules covering EU place of supply, electronically supplied services, OSS route, reverse charge, invoice requirements, customer location evidence, VAT ID verification, and treaty review.
  • Does not decide the tax result; instead, it outputs an adviser brief showing how the transaction was classified, which rules may apply, what information is missing, and what a professional should review.
  • Includes a citation layer that links claims to official sources with footnotes.
  • Has a workflow for answering questions about missing facts, which re-runs the rule engine without another model call.

Evidence

  • The author describes how GPT-5.6 handles five server-side tasks: turning input into typed facts, separating components, generating questions, explaining analysis, and comparing scenarios.
  • The tool uses TypeScript code for tax mapping, not AI alone.
  • It integrates with VIES for VAT ID validation.
  • The application is deployed as a working system.

Inference The product appears to be a prototype or MVP built in a short time frame under constraints (e.g., using Codex, limiting model use to avoid incorrect explanations).

Back to contents

Positioning & Claim Evolution

The author states:

  • They have worked in tax for 15 years and now work independently as a consultant.
  • The problem they identified is that small software companies often cannot afford multiple advisers for every new contract but risk costly VAT mistakes.
  • Most existing tools become useful only after someone has already classified the transaction.
  • TaxGraph aims to help with the step before classification — identifying tax touchpoints, missing facts, and questions.

Evidence

  • The author says they wanted a tool that helps with the pre-classification step.
  • They describe their longer-term idea as a platform guiding small companies through cross-border deals including tax, documentation, certification, and compliance.
  • This project was cut down to one part for the hackathon.

Inference TaxGraph is positioned as a pre-tax decision support tool aimed at small businesses or consultants who need early insight into complex international transactions. It is not described as replacing professional advice but rather as preparing it.

Back to contents

Target Customer & ICP

The description states:

  • The target audience includes small software companies wanting to sell abroad.
  • These companies often lack the resources to hire several advisers for every new contract.
  • The tool is intended to help them understand tax implications before entering a new market.

Evidence

  • The author mentions that small software companies come to them with this problem.
  • They note that VAT mistakes can easily cost more than the advice they were trying to save money on.

Inference

The ICP likely includes:

  • Small-to-medium-sized B2B SaaS or AI service providers selling internationally.
  • Independent tax consultants or firms serving such clients.
  • Possibly early-stage startups entering EU markets.

Back to contents

Business Model & Pricing Evidence

Not evidenced.

The description does not mention any pricing model, monetization strategy, or business model. There is no indication of whether the tool will be sold directly to users, offered as part of a larger platform, or used by consultants.

Back to contents

Technical & Delivery Signals

The author states:

  • The application was built with Codex in one retained development session over three days.
  • The repository contains seven scoped pull requests and 63 tests.
  • GPT-5.6 runs in production for five limited server-side tasks.
  • The system uses TypeScript code for tax rules, not AI alone.
  • It includes a citation layer that checks claims against stored excerpts.
  • The tool rejects claims without valid supporting references.
  • It preserves the source of each fact (form, free text, contract, user answer).
  • The missing-facts workflow updates analysis without re-running the model.

Evidence

  • The build process involved Codex with restrictions: no invention of tax content and no claiming that something worked unless it had run.
  • Tests, builds, commits, and known limitations were recorded in a build log.
  • The system uses schema validation for model outputs.
  • It has a citation gate where claims must reference valid sources.

Inference The technical architecture shows an engineering-heavy approach with AI used primarily for extraction and explanation, not decision-making. The tool is described as inspectable and auditable due to its build log and test coverage.

Back to contents

Traction & Maturity Signals

Not evidenced.

There is no mention of actual users, revenue, customer adoption, or usage metrics beyond the author’s own testing. No evidence of traction, growth, or market validation is provided.

Back to contents

Competitive Context

Not evidenced.

The description does not reference any competitors or existing tools in this space. The author only notes that most tools become useful after classification, but does not name or describe competing solutions.

Back to contents

Key Risks & Red Flags

Risk 1

The tool is described as a prototype built in three days using Codex. It is unclear whether it has been tested with real users or validated in practice.

Risk 2

Although the author says GPT-5.6 does not decide tax outcomes, the system still relies on AI-generated explanations and fact extraction. If those are inaccurate, the tool could mislead users.

Risk 3

The tool is designed to be a pre-tax decision aid, not a replacement for professional review. However, if users treat it as authoritative, this could lead to liability or misuse.

Risk 4

There is no evidence of any commercialization strategy, pricing model, or customer acquisition plan.

Red Flag

No evidence of real-world usage, revenue, or traction — only self-reported development and testing.

Back to contents

Diligence Questions To Ask The Founders

  1. Has the tool been tested with actual users or clients beyond the author?
  2. What is the current level of accuracy in fact extraction and rule application?
  3. Are there any known limitations or edge cases that have not yet been addressed?
  4. How does the tool handle changes in tax law or jurisdictional updates?
  5. Is there a plan for scaling beyond EU markets, and how will local rules be incorporated?
  6. What is the intended monetization strategy, if any?
  7. Are there plans to integrate with legal or accounting platforms?

Back to contents

Investment/Partnership Verdict

Not evidenced.

There is no evidence of funding rounds, valuation, headcount, or investor interest. The project is described as a hackathon submission by one person and lacks any signs of commercial traction or institutional support.

The tool appears to be a functional prototype with clear engineering discipline but no demonstrated market demand or business model. It may have potential for further development, but there is no basis in the description to assess its viability as an investment or partnership opportunity at this stage.

Back to contents

Source

Submitted to the OpenAI 2026 hackathon on Devpost. Project home on DevPost.

The analysis above was generated by a language model from the project's own one-line description. It is not independent research and contains no verified traction, revenue or customer data.