~ $ mildo — ai engineering studio

LLM & agentic AI systems,
built to production standard.

From discovery to your first product version — delivered project by project. Not slideware. Systems that actually run.

4
Live Products
53
Agent Forecast Pipeline
432K
Knowledge Graph Nodes
10
Languages Served

// service_packages

Project-Based Delivery — 3 Packages

Contract only the stage you need. Each stage starts and ends independently, and its deliverables feed the next one.

Start here
01 1–2 weeks

AI Discovery

We start by defining what to build, why, and in what order. Stakeholder interviews, structured requirements, and an actionable technical roadmap.

  • + Stakeholder interviews & process audit
  • + Structured requirements — draft spec
  • + Architecture & data strategy proposal
  • + Phased roadmap + build estimate
02 2–4 weeks

Prototype

A working proof — not slides. We validate the core hypothesis on your real data and give you the evidence to decide whether to proceed.

  • + Working demo on real data
  • + Hypothesis validation & quality report
  • + LLM cost & latency measurement
  • + MVP scale-up design
03 4–8 weeks

First Product Version (MVP)

An operable v1, delivered. Deployment, monitoring, and documentation included — handed over so your team can run it.

  • + Production deployment — cloud infra
  • + Monitoring & evaluation pipeline
  • + Ops documentation & handoff session
  • + Post-launch stabilization support

Per-stage contracts · Corporate tax invoices (KR) · Remote-first, on-site meetings when needed

// work

Live Products

Not portfolio demos — services we designed, built, and operate in production today. Click through and see for yourself.

01

cropcast.ai

LIVE ↗ Visit

Agricultural price forecasting engine

A multi-agent forecasting system where 53 agents collaborate. Public wholesale-price data is ingested daily, structured into a knowledge graph, and published as debate-refined forecast reports — fully automated.

53
Collaborating agents
Daily
Automated pipeline
KG
Knowledge-graph based
Multi-Agent Knowledge Graph Python LLM Pipeline Automation
02

provenio.art

LIVE ↗ Visit

Art knowledge graph + MCP API

A 432K-node knowledge graph built on the CIDOC CRM ontology. Artwork, artist, and provenance data served over an MCP API — a B2B intelligence service with paying international customers.

432K
Graph nodes
MCP
API interface
B2B
Paying intl. customers
Neo4j CIDOC CRM MCP Server Ontology B2B API
03

subswap.app

LIVE ↗ Visit

Global ingredient-substitution PWA

A global web service for finding ingredient substitutes. An installable PWA in 10 languages, running on a serverless architecture that keeps operating costs near zero.

10
Languages
PWA
Installable web app
$0
Server cost
React TypeScript PWA i18n Serverless
04

Myoun

LIVE ↗ Visit

iOS saju (Four Pillars) analysis app

An iOS app combining a traditional Korean astrology calculation engine with an LLM interpretation pipeline. A custom calendrical engine built on astronomical data ensures accurate charts beneath the AI layer.

iOS
Native app
Engine
Custom calendrical core
LLM
Interpretation pipeline
iOS Swift LLM Calendrical Engine

// expertise

Four Pillars of Our LLM Engineering

Every one of these is running in production in the products above.

{}

Multi-Agent Orchestration

Problems a single prompt can't solve, solved by agents working together — role separation, quality gates, and per-model cost routing.

Claude API OpenAI API LangGraph Agent Loop Quality Gates
(G)

RAG · Knowledge Graphs · Ontology

Beyond document search: we structure domain knowledge as graphs — the data foundation for services that must answer with evidence.

Neo4j CIDOC CRM Vector Search Embedding Cypher
<->

MCP Servers

We turn your internal data and APIs into tools LLMs can use safely — connectable from Claude, ChatGPT, and any standard MCP client.

MCP Tool Use API Design Auth & Permissions
>_

AI Backends · Pipelines

Ingest → process → infer → publish, running automatically every day. Designed for operations: monitoring and failure recovery included.

Python FastAPI ETL Scheduling Observability

// team

Engineering Lead

Criminal-intelligence data analysis → enterprise AI & security consulting → legal-tech AI products. A track record proven in the field, leading every project directly.

10 yrs

Korean National Police Agency / Criminal Intelligence · Data Analysis

Built and ran data-driven investigation and analysis systems handling large-scale unstructured data in intelligence-crime investigations.

Data AnalysisInvestigative Intelligence

3 yrs

IBM / AI & Security Consulting

Delivered AI and security consulting for enterprise clients. Two perfect 10/10 client evaluations.

Enterprise AISecurity ConsultingClient rating 10/10 ×2

1.5 yrs

LBOX (Legal Tech) / Legal AI Product — Build & Operate

Built and operated production LLM products in the legal domain — owning the full retrieval, generation, and evaluation pipeline.

Legal AILLM ProductsProduction Ops

Education

Georgia Tech

MS Analytics (in progress)

Credentials

CISSP CCSP IBM client rating 10/10 ×2

// contact

Not sure where to start with AI?

Send a short note about where you are and what you want to build. We reply within two business days and decide together whether discovery is the right first step.

Per-stage contracts · Corporate tax invoices (KR) · NDA available