Associate-level AI Product Manager who ships AI products end-to-end — spec → prototype → evaluation → iteration → ship — with an eval-first mindset focused on whether the product actually works in real usage. I’m good at (1) Making probabilistic systems shippable: clear acceptance criteria, rubrics, golden datasets, failure-mode thinking; (2) Agentic workflows: defining roles, boundaries, handoffs, escalation rules; measuring reliability (not vibes); (3) Fast, builder-style product execution: translating ambiguity into specs and shipped prototypes
INTERLINK LABS
Owned end-to-end PRD authorship and Figma user flow design for 73+ shipped features across a biometric identity app (iOS + Android), spanning onboarding, identity verification, marketplace, and profile management. Specified product requirements for AI verification failure cases (deepfake bypass attempts, near-identical identity edge cases, SDK readiness), translating probabilistic failure modes into shippable requirements for a NIST FRTE-benchmarked facial recognition system. Applied FAR/FRR tradeoff thinking to define acceptable failure thresholds balancing fraud prevention against user friction. Coordinated collection of 2,000+ edge-case biometric images for model retraining, closing the loop between production failures and model improvement. Drove 10+ iOS/Android releases over 6 months at a bi-weekly cadence; improved release efficiency by 30% via tighter QA scope definition and cross-functional coordination. Designed partner integration framework + in-app onboarding funnels driving 1M+ user activations in 8 weeks via targeted partner activations. Built an AI-powered internal CRM (“Network Map”) end-to-end: collected relationship data into Supabase, built the platform using coding agents, and deployed on Vercel; shipped relationship graph visualization, an AI chatbot for querying ecosystem connections, and auto-enrichment workflows. Co-built CurAgent, an internal multi-agent AI ops system on Telegram: owned specification, training, and evaluation; defined a 3-agent architecture, handoff logic, and output acceptance criteria; wrote eval rubric + success criteria per handoff to measure reliability. Served as PM coordinator for CURATED (IF 2.0), an on-chain deal infrastructure platform: owned centralized tracker, weekly objectives, and cross-functional alignment across product/tech/legal/smart contracts; designed and ran an end-to-end mock sale across 4 scenario types with 15 internal participants. Built workflow automations with n8n and Make (meeting-to-Notion, Telegram-to-Notion pipelines, data collection flows), reducing manual coordination overhead across concurrent initiatives.
IMPOSSIBLE
Supported end-to-end research across 10+ enterprise software products; conducted 44 interviews across 3 projects; synthesized insights into product + GTM recommendations. Supported market entry research across 5 markets (4 international) for a B2B loan management platform: competitive landscape, regulatory constraints, and market dynamics → data-driven GTM strategy inputs. Designed and ran a quantitative survey for a mutual fund trading platform (n=200) across Hanoi and HCMC with stratified demographic quotas via Toluna. Contributed to 8 research proposals using mixed methods (IDI, FGD, quant surveys) and structured sampling plans.
FPT IS
Researched topics across 7 tech industries; analyzed 20–30 competitors per industry to shape insights for 9 B2B tech and finance event agendas. Coordinated 20+ conference sessions and managed ~50 speakers; produced 15 marketing material packages contributing to 30% attendance growth.
IEC GROUP
Bachelor
Lightweight evaluation framework for AI products/agents emphasizing “analyze failures before measuring” (trace review, failure clustering, rubrics, golden datasets).
AI-native operating system replacing a traditional product team via a knowledge hub + agent roster + execution stack; runs a repeatable loop from discovery to ship with human approval gates.
AI-native startup validation workbench that turns vague ideas into testable assumptions, evidence thresholds, and decision-grade validation briefs. Built with a strict spec → prototype → eval → changelog → re-eval loop; optimized to reduce false confidence rather than generate prettier plans. Designed an eval framework before prototyping: 8 builder archetypes × 5 graded dimensions (golden dataset). Ran early evals on Gemini models to validate prompt quality pre-infrastructure; iterated to a locked v3 spec.
Relationship intelligence product mapping hidden connections using graph visualization, RAG, and automated enrichment agents.
Founded Vietnam’s AI agent builder community chapter; secured sponsors (Google Cloud, Alibaba/Qwen, AWS/Kiro, Genesia Ventures). Delivered Meetup #1 with 200+ registrations and 93% return intent; iterated strategy based on post-event audience analysis.