About
Execute. Done.
An AI Business Operating OS for solo founders.
Costart builds an L1 AI Business Operating OS, applies it to our own business, validates it on real revenue, and hardens it through verticals, teaching, and consulting.
6 live products · 4 languages · one-founder operations
STAR-T’s journey for local revitalization →Not claims. URLs.
Every one of these opens live, right now. Only dogfood-passed products are sold externally.
AI executes. Humans decide.
Help solo founders / builders full-stack, at every stage from idea to lifestyle, in the AI era.
The first Korean reference at the AI × XR intersection, and the backbone of the solo founder community.
Verify, accumulate, and own AI output. Source, hallucination flags, and audit logs are baseline.
No shallow wrappers. Build at least two moats per domain.
If the founder ain does not use it on their own work, we do not sell it.
We name what we cannot do (e.g. Korea-only tax scope, LLM accuracy limits).
AI works, ain decides. Six-hour daily operating rule.
Models · Data · Infra.
- GPT-5 · Claude Opus / Sonnet · Gemini 2.5
- Gemma 3 4B · local fallback (Ollama)
- OpenAI text-embedding-3-small
- Router: cloud / hybrid / local
- PostgreSQL + pgvector (IVFFlat)
- tax_ai RAG in production · statute OPEN API auto-ingestion
- Append-only ledger
- decisions · executions · outcomes · source_claims
- Next.js 16 · App Router · 4-lang i18n
- Vercel Fluid Compute · Cron · MCP
- GitHub Actions · typecheck · test gate
- Observability: GA4 · source_claims · KPI cron
L1 / L2 / L3 / D.
Every product is built and operated on the L1 OS.
AI business operating OS for solo founders. Threads dev, funding, marketing, ops, tax, QA, analytics.
L1 principles applied to a founder’s endurance: body, food, motivation, memory, relationships.
External vertical AI products, actually built and validated on L1. Only dogfood-passed products ship.
Discovery candidates testing untapped vertical AI markets on top of L1.
A 24-step journey, 12 service shells.
| Shell | What it does | Journey steps | Status |
|---|---|---|---|
| 01 AI MVP Builder Problem validation & product planning | Idea in → PRD, MVP scope, screen/API/DB drafts, dev tasks, QA plan | 1 · 2 · 3 · 5 | Contractable |
| 02 Funding / RFP OS Funding & RFP | Upload a call or RFP → fit, proposal, business plan, submission package, send approval | 7 · 21 · 22 | Contractable |
| 03 Marketing OS Marketing & GTM | Positioning → funnel → channel strategy → content queue → performance report | 14 | Contractable |
| 04 Auto Blog / SEO OS Content & SEO | Keywords → SEO articles → publishing plan → traffic report | 15 | Contractable |
| 05 CS/CX OS Support & VOC | Collect inquiries → auto-tagging → VOC summary → reply drafts → improvement backlog | 17 | Contractable |
| 06 Business AI OS Core Orchestration & governance | Route requests → run → QA → approve → canonize → log | 10 | Contractable |
| 07 Sales CRM OS Sales, partnerships & CRM | Register a lead → deal brief → proposal → follow-up → handoff | 13 · 16 | Contractable |
| 08 Revenue / Billing OS Revenue, billing & settlement | A live board for revenue, billing, MRR, settlement and evidence | 9 · 19 | Contractable |
| 09 Compliance Gate OS Quality, risk & release | Approve or block external sends, personal data, security and overstated claims | 8 · 11 · 12 | Contractable |
| 10 Analytics OS Performance & data | KPIs, funnels, usage, cost, data gaps and performance reports | 18 | Contractable |
| 11 Brand / Design OS Brand & design | Name, logo and brand tone → detail pages and introduction assets | 6 | In design |
| 12 Exit / Closure OS Closure & restart | Decide on closure or pause → procedures, tax and asset wind-down → a path back | 24 | In design |
The 12 shells are contractable units of sale (design scope). Revenue today comes from lectures and consulting and from what we sell through Modu-Startup (AI MVP Builder, BIDS); TaxPro-KR is pre-revenue. Shells 11 and 12 are in design and not yet sold. Journey steps refer to our 24-step journey.
Design → apply → verify → re-harden.
One founder + an AI agent team.
An 11-agent system is designed; the core agents run daily while the rest roll out in stages. The system does not stop when the founder steps away — a 07:30 operations brief ships every day, sales follow-ups track automatically, proposals draft on demand.
| Codename | Role | Domain | Cadence |
|---|---|---|---|
| tae_ops | Ops Lead | Portfolio operating state | Auto · daily 07:30 |
| soo_research | Research Lead | Market · policy · RFP analysis | On demand · auto |
| yuri_bizdev | Proposal Lead | Auto proposal & business plan drafting | On call ingestion |
| mina_cmo | Growth Lead | Marketing, content, channels | Content queue auto |
| hana_cfo | Finance Lead | Finance, revenue, cost summary | Monthly auto |
| jin_cpo | Tech / Product Lead | Code diagnostics & product strategy | PR-triggered |
| memory_lead | Memory Lead | Info classification & ledger | Daily 23:30 auto |
| customer_lead | CRM Lead | Sales pipeline & follow-up | SLA-trigger auto |
| schedule_lead | EA Lead | Schedule, contacts, checkpoints | Weekly auto |
| research_intern | Research Intern | Primary source collection / crawling | Pipeline-linked |
| archivist_jin | Archivist | Retros & quarterly archives | Quarterly auto |
Build product and operations together.
Every external publication cites only the verified entries in source_claims.jsonl. Unverified items are marked [UNVERIFIED].
You run and approve each step yourself. We only reduce wasted loops through routers, skills, and QA — the final responsibility stays with you.
Requirements → MD Task breakdown → router-enforced stages → no unauthorized entry. This reduces both how often AI spins its wheels and what it costs when it does.
Firmware/hardware development ❌, automated tender curation ❌, medical device software (IEC 62304) ❌, tax law outside Korea ❌, outsourced development ❌.
Beyond how we build, how we work
A one-person company has no senior to learn from by osmosis. So we fix tacit knowledge in documents — and publish them as they are.
We separate what we verified, what we heard, and what we judge. Speaking on intuition carries matching responsibility.
No practice assignments. After review, it goes to real customers.
Stuck for 30 minutes? Speak up. Holding it alone is not diligence.
Even if the founder works that way, it is not the standard. Urgent work waits for the next working day.
2020 — 2026, cumulative.
- 135
- Lectures & consulting
- Repeated engagements, not one-off talks
- 1,000+
- Total learners
- Educational design across a wide range of skill levels
- 35
- Core B2B programs
- 2020–2026 accumulated curriculum operation
- 1,100+
- Material & session evidence
- Structured material-driven sessions, not improvised
- 4.71/5.0
- Satisfaction
- Self-run surveys, tied to 1,100+ pieces of session evidence
- 4.68/5.0
- Recommendation
- Self-run surveys — repeated field-applicability feedback
- 4 consecutive years with Shinhan Financial
- Selected as a Modu Startup AI service provider
- B2G PoC with Eumseong-gun; Busan Global Fintech in progress
- tax_ai live: Korean-tax RAG (pgvector + statute API)
Teaching, consulting, collaborating.
Organizations we have taught, consulted, or collaborated with — including four consecutive years with Shinhan Financial Group.
2020 — 2026.
- 2020Lectures & consulting begin
Corporate lectures and government grant mentoring begin — start of the 135-engagement run.
- 2021–2022XR pilots / fintech build
XR pilots in Incheon and Seoul tourism. Fintech v1 with FSI / KFTC certification experience.
- 2023–2024Vertical builders accumulate
Vertical dogfooding starts: ko_living_ai (foreigner settlement), tax_ai (tax-law RAG), jodal (funding). 4 consecutive years partnering with Shinhan Financial.
- 2025AI OS in earnest / 25 products
L1 AI business operating OS hardens. 28 AI products (12 live) across 10 sectors run by one founder. Selected for Modu Startup’s AI service provider listing.
- 20269 businesses · AI agent system · B2G PoC
11-agent system designed with core agents running daily (07:30 brief). B2G PoC with Eumseong-gun (Chungbuk) and Busan Global Fintech in progress. invest (research), xr (XR Agent Builder), and care (CareSync) went live.
Founder
박정근 (ain)
Park Jeong-geun (ain). Fintech v1 founding with FSI/KFTC certification, XR pilots in Incheon and Seoul tourism, four consecutive years partnering with Shinhan Financial. Runs STAR-T solo on a six-hour-per-day discipline.
FAQ
- Do you take outsourced development?
- No. STAR-T is not an outsourcing dev shop. Our product line (e.g. AI-Developer-KIT) is designed so the founder ships the product themselves.
- How much teaching and consulting have you done?
- From 2020 through 2026: 135 engagements, 1,000+ learners, 1,100+ pieces of teaching material, satisfaction 4.71 and recommendation 4.68 out of 5.0, and four consecutive years partnering with Shinhan Financial.
- How big is the team?
- One human founder plus an AI agent team. An 11-agent system is designed; the core agents (operations brief, research, proposal drafting) run daily, and the rest roll out in stages. The 07:30 operations brief, automatic sales follow-ups, and on-demand proposal drafts continue even when the founder steps away.
- What do you not do?
- We do not build firmware or hardware directly, we do not auto-curate public funding calls, and we do not handle medical-device SW (IEC 62304) certification. We name our limits — for example, tax scope is Korea-only.
Costart Inc. (STAR-T) · Founder Park Jeong-geun · Incorporated 2024 (active since 2020)
No upsell — only what fits.
Get More Benefits
Don't miss the latest information by subscribing to our newsletter and joining our KakaoTalk group
Newsletter Subscription
Receive the latest IT trends and insights
Join KakaoTalk Group
Real-time communication and networking opportunities!
Participation Benefits
- ⭐Weekly sharing of latest industry trends
- ⭐Free lecture materials download
- ⭐Networking with other entrepreneurs
- ⭐Priority notification of new STAR-T services
Please leave a greeting after joining!