ABOUT US
We fuse AI agents and automation into every step — planning, design, development, and operations — and deliver faster, higher-quality work than a web agency.
76+
cumulative repos · projects
4 yrs
full-stack + AI-engine delivery
15+
B2B and public-sector references
9
patents filed (child language / voice analysis)
Our CEO and internal director are both rooted at KAIST. The bridge between academia and engineering is where we start.
Chief Executive Officer
Internal Director
TEAM
Planning, design, AI, backend, frontend, mobile, deployment — one team. Lower communication overhead, lower quality risk.
Domain research, prompt / pipeline design, even patent specs — we design the engine and the UX side by side.
Next.js 14–16, React 19, Server Actions, RLS, PL/pgSQL, Cloud Run — all in live production.
Expo SDK 55 New Architecture in production, offline-first Kotlin Android, real field operations.
Letter from the founder
Hello,
Song & Stark is a full-stack technology firm that closes the full loop — product, AI, full-stack, mobile, deployment — inside one team. Our core engineering sits in Seoul, and Song & Stark Rwanda Ltd. engineers in Kigali share the same codebase. An in-house AI translation and summarization pipeline lets both offices operate fully in Korean, and with no sales PM in between, the distance between a decision and the code stays short.
First, we built the AI we wanted to use. SpeechMap fuses post-STT linguistic analysis with the raw acoustic and prosodic features of the voice, inside a single engine. We have filed 9 in-house patents in child-language development and voice analysis, and the clinical design is reviewed in partnership with Dururbarun Social Cooperative. Earlier, we built the Gangwon Provincial Office of Education’s early-childhood language screening system — that on-the-ground experience became the foundation for SpeechMap. Because the engine was designed from the seat of the people who actually use the test, the same instincts carry over when we wire AI into your product.
I studied law and business, then ran IT startups for more than ten years and taught myself to engineer along the way — Song & Stark is my fifth venture. The "Stark" in the company name comes from my co-founder, whom I met in 2021 through KAIST’s Social Entrepreneur MBA. In October 2025 I put down, in a book titled Singularity Inside (Goodspell Design), how AI is rewriting the way we live and build — in part to pin down, for myself, where the engines we build ought to go. Since 2025 I also serve as an Adjunct Professor at Seoul Media Institute of Technology, designing startup-education systems. The phrase we quietly keep — "a technology company on the right side" — starts there: the AI we build should land, in the end, on the right side of one real person.
Whether you really need AI in your product, what shape it should take if you do, or whether this particular turn should skip AI and just ship product fast — just send the question. No sales PM between us: I or a core engineer replies directly, within one business day.
Song & Stark started as a college-student social enterprise on Rwandan soil twelve years ago. That arc has closed a full cycle.
2011
CEO Song Wan, then an undergraduate, co-founds the KOICA-backed social enterprise Rz MANNA in Rwanda. The two years he spent running a 30-person local team became the seed for what later grew into Song & Stark Rwanda Ltd. and our “technology with social value” core.
2012–2014
After serving as Rz MANNA’s General Manager (2012–2013), Song Wan led grant-writing for KOICA public–private cooperation, securing ≈ $1M over two years. In 2013 he supported Transform Africa Summit on the ground and served as on-call English interpreter for MWC (at YSK Media). In 2014 he co-founded Next Aura (crowd-delivery Packmanz · African ad platform EVERIKA), pivoting fully into IT startups — and won the Ewha Startup Competition Grand Prize and the Korea Startup League president’s award.
2015–2020
Song Wan founded People-Plus Inc. in 2015 and ran it through 2023. The company shipped the “Syndipost” people-finding app and the “Moim-과정모” group-management service, led two government Startup Growth R&D grants, and won the 11th Korea Service Satisfaction Awards (e-Service / Social App). Selections: Smart Venture Startup School (2015), SMBA Global Startup Program (2016), Seoul Startup Hub Early-Stage Program (2017).
2018
Song Wan served as an adjunct researcher at KAIST’s Advanced Defense Convergence Research Center, working on open-source intelligence (OSINT). The KAIST relationship formed here would later run through the Social Entrepreneurship MBA and the current internal-director seat.
2021
Song Wan entered KAIST’s Social Entrepreneurship MBA (co-founded by KAIST and the SK Group), where he met the current co-founder. In the same year he also served as CEO of Book Band Inc., running a book-sharing platform.
2022
Seoul HQ launched in Gangnam. A law-and-business foundation plus more than ten years of IT startups and self-taught engineering, rebundled for the AI era — the CEO’s fifth venture. The “Stark” in the company name comes from the co-founder met through the KAIST SE MBA.
2023
The dedicated engineering subsidiary in Kigali was incorporated on 2023-02-08. Twelve years after Rz MANNA, the Rwandan thread returned — this time as an engineering entity. Top in-country engineers now work inside the same codebase as Seoul HQ. The same year, Song Wan served as a startup coach at the Apple Developer Academy @ POSTECH, and the company was selected into Logistics Voucher, Non-Face-to-Face Service Voucher, Re-Challenge Success Package, and Korea–Africa EdTech Open Innovation (Stage 1).
2024
Selected into the Africa Frontier Acceleration Program (2nd cohort), KOSME Young Founder Startup Fund, and the Domestic-to-Export Transition Program — expanding both domestic and international business foundations.
2025
We built the province-wide early-childhood language screening system for the Gangwon Provincial Office of Education, reaching 3,500+ children — the field experience that would become the foundation for the SpeechMap engine. Selected into SeoulTech’s Early-Stage Deep-Tech (AI · Big Data) package; Certified Korean venture status; Data Voucher Program delivery; seed funding closed. CEO Song Wan was appointed Adjunct Professor at Seoul Media Institute of Technology (BM Lab · startup education), and in October published Singularity Inside — The World AI Changes and Our Future (Goodspell Design).
2026
SpeechMap officially launches and goes live inside Dururbarun Social Cooperative’s speech therapy program. Korea System Certification ISO 9001:2015 Quality Management System (Q-CA) obtained. 9 in-house patents filed across child-language development and voice analysis.
The official credentials that sit alongside what we’ve built.
Scans of the official documents we can display. Click a thumbnail to see the full-resolution version.
Korea System Certification (KSC · KAB-QC-55)
Scope: Development and Supply of AI Voice Analysis-Based Software
Korea Technology Finance Corp. (KOTEC)
Scope: Venture investment track
Our in-house patent portfolio of 9 filings in child-language development and AI voice analysis — two prior filings, five new 2026 filings, plus two earlier engine / clinical-reporting filings.
Decomposes a child’s utterance into 2D phoneme units and auto-scores PCC via dynamic programming weighted with clinical error severity — replacing labor-intensive manual scoring.
Suppresses LLM hallucination and pairs it with hybrid acoustic forced alignment to predict clinical speech indicators — designed specifically for the high STT error rate of early-childhood speech.
Extracts five language indicators from multi-task elicitation (picture description, sentence repetition, counting) in roughly three minutes and converts them into T-scores from standardised norms — designed for direct operation by front-line teachers.
Identifies deficit patterns in screening scores and uses regression models to automatically recommend which standardised follow-up test (APAC, REVT, PRES, etc.) to administer next.
Accumulates repeated-screening scores into a time series, analyses slope and inflection, and classifies cases as responders / non-responders / plateau — evidence that speech therapy is actually working.
Auto-matches clinical speech records, gates them through a quality check, and retrains prediction models with passing data — an engine that gets smarter over time on its own field data.
Orchestrates the full screening → scoring → deficit identification → recommendation → standardised test linkage → training data → retraining → engine improvement loop as a single platform.
An AI engine that fuses post-STT linguistic analysis with prosody-level acoustic features to flag early language-development risk in young children.
Generates clinician-reviewable assessment reports and closes the loop between guardian feedback and center ERP so intervention stays continuous.
TRUST
We’ve operated on public and medical data — security and IP are enforced at design time, not after.
Personal · medical data
Real consent / collection / deletion workflows for Dururbarun Social Cooperative, Band Foundation, public education authorities, etc.
Supabase RLS
Row-Level Security policies with 12+ role systems across institution / role / level.
Cloud Run + Artifact Registry
Container-isolated engines, logging and rollback, 4GB/2vCPU production.
Accessibility
We built a11y-checker, our own KWCAG auto-audit SaaS; deliverables ship to AA.
NDA · copyright
Client NDAs, IP assignment, open-source license audits.
Patents
9 filed
In-house patent portfolio in child-language development and voice analysis
Repositories
76+
2022–2026 cumulative repositories, 42 shipped products
Live Products
15+
Currently operating — ours and client services
COLLABORATION
What happens after your inquiry, in the open. Biweekly demos turn decisions visible.
W 00
Goals, scope, timeline locked in a shared document.
W 01–02
Users, competitors, data — summarized and shared.
W 03–06
IA, wireframes, design system locked.
W 07–14
Biweekly demo cadence.
W 15–16
Devices, browsers, accessibility verified.
W 17–∞
Monthly reports and improvement sprints.
The Rwanda origin that started with Rz MANNA in 2011 came back in 2023 as Song & Stark Rwanda Ltd. (est. 2023-02-08). Some of the country’s best engineers sit inside our same codebase today, carrying AI pipelines and production infrastructure with us.
Two adjacent experiments run alongside our core work. Stark Gallery is a physical space for Asian contemporary art; an AI art-valuation platform is in development as a data tool for collectors. Progress updates live in our Insights channel.
