Work
Products, systems, and experiments built around real operational problems.
K-Ledger
A focused operational accounting system built from the needs of a real operating business.
Problem
Operational accounting information can become fragmented across spreadsheets, receipts, messaging, inventory records, receivables, payables, and cash-flow tracking.
Context
K-Ledger originated from practical needs around Kovaa, a real B2B operating business used as a proving ground for internal systems.
System
Current scope is centered on operational accounting and evidence, including receivables/payables, inventory-related visibility, and cash-flow awareness.
Design decision
K-Ledger is intentionally not positioned as a generic full ERP. Tax accounting remains a distinct concern; the system is oriented toward operational clarity and human review.
Prokron
One project. Shared understanding.
Problem
Complex AI-assisted work loses continuity whenever the model, session, or person changes. Decisions, intent, and current position have to be reconstructed from chat logs, and a newcomer has to read implementation code to learn what the project is.
Context
Born from continuous daily development with generative coding agents and multi-session project lifecycles. Carried here under the working name Handoff until September 2026, and now developed in public.
System
Two entry modes: a new project derives its first tasks and decisions from its specification, while an existing one starts empty and records from the current session forward rather than inventing a history. Work then runs one task at a time — resume, work, decide, checkpoint. Installation is a single shell command, and the workflows are portable Markdown, so Claude Code, Codex, OpenCode, and any agent that reads an instructions file follow the same agreement.
Design decision
Prokron is records and workflow, not a runtime — Markdown files, agent instructions, and portable command prompts, with Git keeping file history. Completion and validation stay separate: a task is done when its acceptance evidence exists, and its validation is recorded as untested, synthetic, AI-reviewed, or human-verified, which only a named human may set. A changed decision is superseded by a new record; earlier ones are never rewritten.
LeviDraft
A structured environment for bilingual legal drafting and review.
Problem
Bilingual legal drafting is not simply a translation problem. Source provenance, legacy language, drafting choices, review, and reconciliation all matter.
Context
Originates from real bilingual commercial and regulatory contract workflows in Vietnam.
System
Current exploration includes provenance tracking, legacy bilingual mining, structured drafting workflows, reviewer reconciliation, and evaluation/reward mechanisms.
Design decision
Do not position LeviDraft as “a better translator”. The product thesis is a specialized legal drafting and review environment.
We prefer systems that emerge from real constraints over technology built in search of a problem.