_@baodev

I build backend and AI systems for real operational use — explicit state, bounded decisions, and failure-aware automation.

_{Long-form notes on AI systems, agents, research, products, and the engineering decisions behind them.}

Most of my work sits at the intersection of backend systems, agent systems, and applied research. I build systems that have to operate under uncertainty: durable state, explicit routing, bounded model authority, evidence-aware decisions, and recovery when the happy path breaks. I also design Agent Skills and knowledge architectures that turn domain research into usable, testable decision systems rather than static prompt or document collections.

Area	Stack
Languages & Runtime
Backend & Security
Data & Processing
AI & Agent Systems
Agent Skill Engineering
Research & Evaluation
Infra & Storage

An open-source Agent Skill for moving from customer evidence to marketing decisions and execution without losing adopted state, evidence boundaries, or the distinction between what is known and what is only hypothesized.

• Decision and work-state discipline — frames the current job, preserves approved choices across multi-step work, and reopens decisions only when new evidence or unresolved dependencies actually require it
• JIT specialist knowledge — routes agents to the smallest useful guidance across customer research, positioning, content, landing pages, email, search/discovery, commerce, paid media, commercial design, brand identity, and scoped localization
• Evidence and claim control — keeps observation, interpretation, attribution, causality, commercial state, platform signals, local evidence, and product truth distinct so fluent copy cannot silently become stronger than its proof
• Falsification-first research — architecture changes go through explicit research, adversarial freeze and implementation reviews, and bounded repairs; candidate chapters, specialists, or primitives are rejected when existing ownership can represent the decision without material loss
• Evaluation with explicit limits — routing checks, behavioral harnesses, and stateful work-episode evaluation are used to test concrete failure hypotheses while recording unresolved, redundant, or non-advantage results instead of converting test passes into universal claims

_{Agent Skill · Python tooling · JIT Knowledge Routing · Marketing Research · Evaluation · Open Source}  ·  Building in public  ·  🌐 MIT-licensed open source

An open-source Agent Skill for founders and operators working through Vietnamese business-law decisions, separating durable legal reasoning from rules that must be verified against current or historically applicable authority.

• Business-first legal reasoning — reconstructs material facts, dates, legal propositions, and decision dependencies before jumping from the user's label to a legal conclusion
• Accountable routing and composition — routes work across BL1–BL8 with one accountable owner per material proposition so corporate, contract, dispute, tax, employment, regulatory, and cross-border analyses can compose without silently overwriting each other
• Live-law authority verification — separates discovery from verification, locks document identity and lifecycle, checks effective periods and amendments, resolves controlling provisions in context, and preserves source drift or unresolved currentness instead of treating the first search hit as law
• Action-readiness discipline — keeps legal possibility distinct from readiness to act and returns options, consequences, unresolved facts or authority, evidence needs, and next actions while preserving uncertainty where the record is incomplete

_{Agent Skill · Vietnam Business Law · JIT Legal Routing · Live-law Verification · Legal Research · Open Source}  ·  Early dogfooding  ·  🌐 MIT-licensed open source

A completed full-stack Vietnamese reading and audio platform built around CDN-backed content delivery, realtime features, background processing, and mobile-first UX.

• Static chapter/audio delivery through object storage + CDN while the API serves metadata and application state
• Redis-backed background jobs for batched counters, statistics, reconciliation, cleanup, rewards, and ranking workloads
• Realtime chat/notifications, PWA support, JWT + Google OAuth, rate limiting, and production deployment tooling
• Separate media/content paths for image, JSON, and audio-heavy workloads instead of pushing everything through the application server

_{Vue 3 · TypeScript · Node.js · MySQL · Redis · Cloudflare R2 · Cloudinary}  ·  Completed system  ·  🌐 Public source

An ongoing research track on state integrity in long-lived AI agents: provenance, authority, temporal correctness, durable execution, and the gap between rich runtime traces and governance-relevant interfaces.

• Separates established foundations from agent-specific hypotheses instead of assuming novelty
• Uses falsification-first gates: weak ideas are narrowed or abandoned rather than rescued after the fact
• Current benchmark work studies whether an already-produced action can still be bound to the exact committed external effect occurrence after crash/recovery boundaries
• Freezes claims, implementation, held-out evaluation, and adjudication rules before observing authoritative results

_{Python · Agent Runtimes · Durable Execution · Provenance · Benchmarking}  ·  Research in progress  ·  🔒 Private until release

A paused product/R&D track for finding products with promising source-market signals on 1688 that still face limited competition on Shopee Vietnam.

• Engine pipeline: source crawling → deduplication → translation → Shopee image/market matching → supply enrichment → evidence-backed scoring and decisions
• Opportunity scoring combines demand, visual gap, estimated margin, local-market gap, competition pressure, data confidence, and explicit risk signals
• Separate SaaS consumes a versioned engine contract and provides auth, subscriptions, billing, opportunity exploration, watch/reserve workflows, and market radar

_{Python · Vision / pHash · Next.js · TypeScript · PostgreSQL · payOS}  ·  Paused  ·  🔒 Private source

These are smaller or more inspectable public artifacts that show additional engineering work beyond the primary projects above.

• BAO.OS / Portfolio — interactive portfolio with command routing, multiple RAG modes/personas, Supabase pgvector retrieval, local deterministic embeddings, an OpenAI-compatible API surface, and server-side redaction boundaries; live LLM/RAG is currently disabled on the public deployment.
• CD1-2 — Linux security monitoring lab combining Suricata, Wazuh, deterministic correlation/risk scoring, local IsolationForest signals, evidence-aware explanations, and controlled response policy.
• Audio Ingest — queue-based YouTube audio ingestion pipeline using Redis workers, FFmpeg, object storage, recovery paths, and database synchronization.

_{Open to backend / AI systems roles where reliability actually matters  ·  reach out}