BuilderT

Posted on Aug 13 • Edited on Sep 5

Yomai: Turn Any Messy Input into Structured Outcomes

#api #dataengineering #ai #datascience

Turn messy inputs into structured outcomes with Yomai

If you’ve ever wished your backend, files, or third-party APIs could produce clean, deterministic outputs — and even be accessed via natural language — this is for you.

I’m building Yomai — an AI service-integration platform that transforms any input into structured data guided by your schema/dictionaries. It’s API-first, deploys cloud or on-prem, and can optionally expose your services to NL or MCP when you want conversational access.

Here’s how it works 👇

🛠 Step 1 — Define your schema (“dictionaries”)

Tell Yomai what “good output” looks like. You define fields, types, and rules — Yomai’s AI aligns to your structure, not the other way around.

// desired_output.json (example)
{
  "customer": { "name": "string", "country": "string" },
  "orders": [
    { "id": "string", "total": "number", "placed_at": "date" }
  ]
}

🔌 Step 2 — Configure the destination

Specify where the request or processed data should go — HTTP method, URL, auth token, database connection, or file output — all from a single screen.

🧠 Step 3 — Transform any input into your desired outcome

Yomai takes any input format — XML, JSON, CSV, Excel files, PDFs, logs, API responses, or even plain natural language — and always transforms it into the structured output you defined in your schema/dictionaries.

No matter how messy the source, the result will match your desired outcome, ready for your applications, workflows, or LLM integrations.

✅ Step 4 — Get structured output you can ship

You receive clean, structured data that drops straight into workflows, databases, or APIs. It’s headless-ready and integrates into CI/CD.

🗣️ Step 5 — (Optional) Expose as Natural Language / MCP

When you want conversational access (for users or agents), Yomai can expose your services as NL interfaces and is compatible with the Model Context Protocol (MCP) — so assistants/IDEs can call your capabilities safely. No custom UI required.

What Yomai is great for

Data Integration: unify JSON/PDF/XML/plain text into one deterministic structure
Document Understanding: extract structured fields from invoices, contracts, logs — guided by your schema
API Glue & Conversion: map between APIs that don’t speak the same shape; use schema mappings instead of brittle glue code
Deploy Anywhere: Docker/CLI for on-prem; cloud when you want speed
API-First: integrate programmatically; keep your stack and workflows

Why this is different

Schema/dictionary-first: you define the contract; AI adheres to it
Universal inputs: handle messy, multi-format reality without one-off scripts
Conversation when you need it: flip on NL/MCP to let humans/LLMs use your services without rebuilding your backend
Cloud or on-prem: fit security and compliance constraints

Quick example

Input: mixed sources
- PDF: invoice.pdf
- API: GET https://payments.example.com/v1/txns?since=2025-08-01
- Text: “Customer Acme placed two orders last Friday…”

Schema: 
→ customer {name, country}
→ orders [{id, total, placed_at}]

Yomai:
- Parses PDF text + API JSON + free-text
- Normalizes dates/currency
- Emits structured JSON that matches your schema

🚀 Try it

Start here → yomai.io

I’d love to hear your feedback — especially your weirdest data edge cases or APIs that refuse to play nice.

DEV Community