Jason Phan

SYS://jason-phan · portfolio

hey, I'm Jason

Full-Stack / AI Engineer

Building agent evaluation and orchestration systems for regulated banking and frontier-model teams.

AI OrchestrationCloud LLM and Document AIBackendFrontendCloud and Data

View projects Download CV Get in touch

agent online · grounded on Jason's CV

ask me anything about Jason's work, stack, or availability.

proc / at a glance

Quick summary

years in production delivery

countries delivered across (VN / US / SG)

companies & banking clients

AI Orchestration

LangGraphLiteLLMPineconeQdrantLangSmith

Cloud LLM and Document AI

Azure OpenAIAWS BedrockOpenAI API

Backend

FastAPIPythonNode.jsDocker

Frontend

Next.jsReactTypeScriptTailwind CSS

Cloud and Data

PostgreSQLAWS LambdaAzure Blob StorageDynamoDB

Full experience Full skill set

proc / clients

Companies & clients I've worked with

FinopticsShinhanKDBTD BankCitibankScale AIEntanaFinopticsShinhanKDBTD BankCitibankScale AIEntana

proc / pipeline

How the eval agent runs

A live-style view of the layered architecture behind Apron, the LangGraph evaluation harness. Drag to pan, use the controls to zoom — the green node shows where a request currently is.

Interface

Next.js UI submits eval runs to a FastAPI gateway

Next.jsFastAPI

live · handling requests

Control plane

LangGraph router dispatches to the skill registry

LangGraph

routing to skill

Skill execution

Python task/language skills run inside a sandboxed harness

Python

running skill

Data & ops

Postgres persists runs; LangSmith traces every call

PostgreSQLLangSmith

persisted · traced

Evaluation & output

Verification

Deterministic checks, then an LLM judge via LiteLLM

LiteLLMAzure OpenAIAWS Bedrock

scoring output

Governance

Rubric scoring, human review gate, then export

gate passed

proc / projects

Selected work

Dec 2025 → Present

Apron — Vietnamese Locale LLM Eval Harness

Built LangGraph evaluation workflows for Vietnamese coding-locale benchmarks with stateful routing, pass/fail gates, and regression batches for instruction-following and tool-use scenarios.

LangGraphFastAPINext.jsTanStack Query

Jan 2025 → Nov 2025

Hopper — Vietnamese Locale Benchmark Fixtures

Built Python ingestion and normalization pipelines that turn industry code samples into structured benchmark fixture libraries for Vietnamese coding-locale eval runs.

PythonFastAPIschema validationfixture libraries

Dec 2023 → Nov 2024

AI Document Translation Dashboard — Korean Development Bank

Benchmarked Korean document OCR across Amazon Textract and Azure Document Intelligence on pilot corpora in Amazon S3 before productionizing extract on Azure VNet.

LiteLLMAzure OpenAIPineconeAzure Document Intelligence

View all projects

proc / contact

Let's talk

Jason Phan

Based in Vietnam. Open to full-stack and AI engineering roles — ask the console above, or reach me directly.

Email GitHub LinkedIn