We turn focused product work into reviewable engineering experience. A person owns the decision, AI agents accelerate bounded work, and evidence determines what we can claim. The studio is a Cambodian student-run product studio, and most product source stays private.
We are intentionally AI-native: agents help us research, build, test, document, and review, while humans stay accountable for product judgment, security, verification, and final decisions.
Most product source remains private. This profile exposes only approved direction, operating standards, and public proof—not customer data, private source, credentials, access state, or strategy.
Chomkar Live pre-pilot |
PHSAROS Live core |
BayonHub Live on sample data |
Svaeng Yul Private preview |
Sat Digital Guard deployed |
Vantrex Pre-launch |
LMS for Education Public demo |
LEAP Everlasting Blooms Client build, live |
Current Product Tracks — the evidence boundary for every card above
These labels describe the evidence boundary as it exists now. They do not imply traction, legal ownership, production readiness, or a public release.
| Track | Direction | Evidence boundary |
|---|---|---|
| BayonHub | Opportunity coordination for Cambodian tech students. | Live at bayonhub.com on fictional sample data, labelled Sample and restored daily: students register, save and apply; employers post and move applications. No real employers, openings, placements, or users yet. Email delivery, admin two-factor sign-in, password reset, and avatar uploads are not in place. |
| Svaeng Yul | Mobile-first QCM practice for Cambodian medical and nursing students. | Restricted private medical QCM preview. Instructor review plus mobile and authentication validation remain open; it is not clinical guidance or a substitute for formal assessment. |
| Chomkar | B2B agriculture coordination where buyer requests can be documented and reviewed with farmers and cooperatives before harvest. | Live pre-pilot; no real order, validated demand, guaranteed price, sale, or impact claim. |
| Sat Digital | Defensive moderation for Telegram communities, groups, and websites. | The rules run as a deployed webhook and act only where a human has made the bot a group admin. No pilot, users, revenue, monitoring coverage, autonomous action, or protection guarantee. |
| Vantrex | Trading decision-support for signals and indicator workflows. | Pre-launch: no customers, checkout not wired, no published track record. Sign-in answers again on a recreated database. Neither a deployment nor a portal establishes validated model performance or live-money capability. |
| PHSAROS | Point-of-sale, inventory, customer, and expense workflows for local SMEs. | A public product surface exists; fresh operator validation and written attribution remain open. No audited accounting, tax, payroll, or compliance claim. |
| LMS for Education | Course delivery, assessment, and academic operations for schools and training centres. | A public demo on synthetic data carries the enrol-to-grade loop. No institution runs on it; production providers, launch security, and approval for live student data remain open. |
Client work — what belongs to the client
The studio also builds for clients, and the brand, product line, and photography in that work belong to the client, not to us.
| Client build | What it is | Boundary |
|---|---|---|
| LEAP Everlasting Blooms | A storefront for a Phnom Penh handmade-gift maker who had been selling through chat messages: catalogue, accounts, checkout, KHQR payment, and order tracking. | Delivered and live. No order volume, revenue, or customer count is claimed here. |
| Principle | Practice |
|---|---|
| Evidence before status | We separate deployed, local, simulated, degraded, and planned work. |
| Humans remain accountable | A person owns product judgment, verification, review, and the decision to ship. |
| AI stays bounded | Agents receive scoped context, never secrets, and their work remains reviewable. |
| Private by default | Source, credentials, user data, financial logic, and unreleased strategy stay out of public surfaces. |
| Credit is explicit | Contribution and product ownership are recorded decisions, not implications created by a profile or repository location. |
mainstays deployable.- Every meaningful change traces to an issue and pull request.
- Pull requests record human verification, AI involvement, risk, and rollback.
- Agents receive context, never secrets.
- Security-sensitive work stays private and follows the org security policy.
AGENTS.md— the vendor-neutral contract for AI agents in CHNAI LAB repositories.docs/AI_AGENT_WORKFLOW.md— the traceable issue-to-merge workflow for humans and agents.docs/REPOSITORY_STANDARD.md— adoption contract for every active product repository.docs/AI_CONCIERGE_STANDARD.md— public, product, and private agent boundaries.GOVERNANCE.md— roles, access, decision rights, and the current enforcement boundary.CONTRIBUTING.md— the path from a ready issue to a reviewed pull request.SECURITY.md— public/private reporting boundary.
ai-native-team-starter is a private product-neutral repository template
prepared for owner review and a separate publication decision.
It is not currently available for outside adoption. Any future public release
must first pass the repository's public-boundary and security review.
Live membership, invitations, security state, roles, and repository access are owner-managed on GitHub and verified privately. This public profile describes the operating model, not a current roster or access inventory.
If a repository is public, it should be safe for a recruiter, partner, student, or automated reviewer to inspect. Public content should show our standards, direction, and proof of work without overstating traction or leaking startup assets.
Browse the current public evidence and case studies at www.kavatana.me/projects.
