Antonio solves the widespread issue of verbal subcontracting in Korea, which accounts for 9.3% of all B2B transactions. Most SMEs (97%) operate without dedicated legal teams, relying on chat messages or phone calls for orders. When disputes arise, they face:

Lack of evidence: Nearly impossible to prove original contract terms

High costs: 19.7% abandon disputes due to legal fees

Legal violations: 29.2% of subcontracts break the Fair Subcontracting Act
Antonio makes the process safer and easier:

Easier: Simply upload a chat screenshot or audio recording of the negotiation

Safer: Our AI pipeline extracts key terms, drafts a compliant standard contract, and flags unfair clauses before signing. The contract hash is then stored on the Base L2 blockchain, providing immutable, verifiable proof.

The biggest challenge was getting the LLM to reliably extract structured contract terms from chaotic chat messages.

The KT Mi:dm 2.0 Base model struggled to consistently output valid JSON, often mixing conversational text or hallucinating formats, which broke our pipeline.

Solution: We defined a strict Pydantic model for contract terms, converted it into a GBNF grammar, and used llama-cpp-python to force the LLM output to strictly conform to our schema—achieving 100% valid and parseable results.
Another major hurdle was the complexity of the AI pipeline. A single, monolithic API route for OCR, extraction, and RAG verification was slow and prone to failure. We re-architected this into separate containerized Kubeflow Pipeline components (OCR, Extract, Verify), creating a robust, scalable workflow on Kubernetes.

https://github.com/kyne0127/antonio

https://github.com/kyne0127/antonio

Antonio’s unique value is transforming unstructured chat and audio conversations directly into legally-sound, verifiable digital contracts.

• Existing e-contract services (like DocuSign) only handle signing of pre-made documents. Antonio solves the “zero-to-one” problem—creating the contract itself from real-world negotiations.

• Our alpha build demonstrates this: users upload chat screenshots, the Kubeflow ML pipeline is triggered, Mi:dm 2.0 extracts key terms, and a compliant contract draft is instantly generated and checked for legal issues.

• Unlike competitors who react after disputes occur, Antonio proactively prevents contract risks from the start.

Our ideal users are SMEs in Korea—typically CEOs, project managers, and sales leads in manufacturing, IT, and construction—who lack in-house legal teams (97% of SMEs).

We validated this audience using government data: FTC found 9.3% verbal contracts and 29.2% violations; Ministry of SMEs reports 19.7% abandon disputes due to cost/evidence. These stats prove our market is large, underserved, and clearly needs our solution.

1. Traditional e-signature platforms (Modusign, Kontrak, DocuSign):

Only handle signing of existing documents—Antonio creates contracts from scratch using AI and real-world evidence.

2. Traditional legal services (Law firms, LawTalk):

Expensive, slow, and reactive—Antonio is cheap, instant, and proactive, acting as an AI legal assistant for real-time compliant contract drafts and risk checks. Blockchain verification provides trust beyond what law firms can offer.
Main competitors:

• https://www.modusign.com/
• https://www.kontrak.co.kr/
• https://www.docusign.com/
• https://www.lawtalk.co.kr/