Danny Pham

Open to remote rolesSenior Fullstack Engineer · Ho Chi Minh City

I build AI products
end to end.

React and Next.js on the front, Node.js and PostgreSQL behind it, and the LLM features in between — streaming UIs, retrieval, agents, and the guardrails that make them safe to ship. Currently at Renn Labs.

askWhat has Danny shipped with AI?
  1. rewritequery clarified
  2. retrieve
  3. fuse
  4. generate
  5. verify

Rendered the way Văn Bản Xanh answers. Citations jump to the work.

Selected work

CQ 2.0 assistant

Agent chat UI · Renn Labs

The docked copilot inside an agent-first fundraising platform — it drafts, plans, and acts on the CRM.

  1. message
  2. agent
  3. tool call
  4. card: plan · form · approval
  5. confirm
  6. write
  • Built the chat UI on assistant-ui and the Vercel AI SDK: generative-UI tool cards, slash commands, @mentions, voice dictation, and a per-thread model picker.
  • Enforced confirmation gates on the write path, so the agent can't delete or send anything the user didn't approve in that turn.

Next.js 16React 19assistant-uiVercel AI SDKZustand

Văn Bản Xanh

Personal project · live

Vietnamese legal Q&A over national laws that answers only from quotes it can verify, and refuses when it can't.

  1. question
  2. legal-vocab rewrite
  3. dense + BM25
  4. RRF fusion
  5. generate JSON
  6. verify quotes
  • Built hybrid retrieval over article-level chunks in pgvector; an LLM rewrite of everyday questions into statutory terms, fused with RRF, improved answer accuracy and reduced false refusals on an eval set.
  • Checked every cited quote against its source chunk server-side and refused answers with none verified, then shipped it on a free-tier Neon + Render + Vercel stack with a daily LLM spend cap and CI eval gates.

FastAPIpgvectorBM25Claude HaikuLangfuseNext.js

Kịp

Personal project · live

A daily IT digest in English and Vietnamese, built only from official sources and fact-checked against them.

  1. official sources
  2. rules + dedupe
  3. LLM score + EN/VI summary
  4. verify quotes & numbers
  5. merge same event
  6. per-stack digest
  • Built the daily pipeline: an LLM scores and summarizes each item with structured output, then code checks the evidence is a verbatim quote and every version, CVE, and figure appears in the source — one retry, then fall back to the original excerpt — and a second model flags headlines it can't verify.
  • Shipped a Next.js BFF, NestJS API with shared zod contracts, OAuth, and Free/Pro plans in a Turborepo monorepo, with a swappable Anthropic/OpenAI provider and a daily budget cap on free-tier hosting.

Next.js 16NestJSPostgreSQLClaude Haiku/SonnetOpenAIGitHub Actions

AI Gateway

Personal project · deployed

An LLM gateway between apps and model providers. Apps change only their base URL; every request gets routed, budgeted, cached, guarded, and traced.

  1. OpenAI / Anthropic request
  2. auth + rate limit
  3. guardrails
  4. budget
  5. cache + verifier
  6. intent router
  7. fallback chain
  • Built intent routing on a TF-IDF model ported to TypeScript, atomic per-tenant budgets in Redis + Lua, retries, circuit breakers, and streaming guardrails; routing plus caching was evaluated against an all-large-model baseline for cost and answer quality.
  • Measured embedding thresholds failing on Vietnamese, so an LLM verifier confirms semantic cache hits; failover was load-tested through a mock-provider outage with k6, and guardrails were tested against an attack-prompt suite.

TypeScriptHonoPostgreSQL + pgvectorRedisOpenTelemetryLangfusek6

Lịch Âm

Personal project · Zalo bot

A Zalo agent that keeps a family's lunar-calendar events — giỗ, lunar birthdays, Tết — in Google Calendar.

  1. Zalo message
  2. agent
  3. lunar ↔ solar tool
  4. proposal
  5. OK
  6. Google Calendar
  • Kept date math out of the model: conversion tools compute every date, create_event rejects a date that doesn't match its lunar claim, and writes pass an approval gate the bot renders itself, so the model can't skip it.
  • Built an eval harness (adversarial, prompt-injection, metamorphic variants) scored by code assertions, an LLM judge, and human calibration; it gates PRs in CI on pass-rate regressions.

TypeScriptVercel AI SDKOpenAIZalo Bot APIGoogle Calendar API

PetZone

Personal project

A two-sided marketplace where pet owners book stays with accredited pet hotels.

  1. owner app · provider portal
  2. NestJS API
  3. Postgres + RLS
  4. realtime notifications
  • Built the consumer app, provider portal, and API as a Turborepo monorepo with shared UI, types, and config.
  • Isolated each provider's data with Supabase row-level security, and pushed booking updates over realtime subscriptions.

Next.jsNestJSPostgreSQLSupabaseTurborepo

Experience

2023 – now

Senior Fullstack Engineer at Renn Labs

Remote, USA

  • Led Next.js apps from build to production
  • Reduced page load and bundle size
  • Design system adopted across teams
  • Mentored junior engineers, supporting their promotions

2016 – 2022

Senior Software Engineer at Tiki

Ho Chi Minh City

  • Tini App mini-app APIs for merchant partners
  • Storefront migration from jQuery to React/Next.js, with faster Time to Interactive
  • Core Web Vitals work that improved SEO and conversion
  • Team Champion of the Year, 2018

2014 – 2016

Junior Software Engineer at VNG

Ho Chi Minh City

  • Landing pages for flagship games
  • Won the Web Developer Competition, 2015
Frontend
TypeScript, React, Next.js (App Router, RSC), React Native, Tailwind, Shadcn/UI
Backend
Node.js, NestJS, FastAPI, PostgreSQL, Prisma, Supabase, REST, GraphQL
AI
OpenAI, Anthropic, Vercel AI SDK, assistant-ui, LangGraph, RAG, pgvector, Langfuse, MCP
Delivery
Vitest, Playwright, GitHub Actions, Vercel, Turborepo, Inngest, Sentry

Contact

Hiring for a senior fullstack or AI product role?

danny.vandp@gmail.com

English (B2) · Vietnamese (native) · Works async with US and EU teams