Skip to content
View HaoPan036's full-sized avatar

Achievements

Achievement: Pull Sharkx2Achievement: YOLOAchievement: Pair Extraordinaire

Achievements

Achievement: Pull Sharkx2Achievement: YOLOAchievement: Pair Extraordinaire

Highlights

  • Pro

Block or report HaoPan036

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
HaoPan036/README.md

Hao Pan

Applied AI / Full Stack Engineer

I build LLM agents, data agents, RAG systems, AI developer tools, and reliable AI applications — grounded in deterministic guardrails and evaluation rather than prompt-only trust. Currently pursuing a Master of Computing in Applied AI at NTU, Singapore; previously worked in AI/ML education before moving into applied AI engineering.

Selected Work

Data-Agent-Sandbox

A runnable BI data agent: a natural-language question becomes validated read-only SQL, an executed result, a chart, and a reviewable trace. Zod-validated request/response contracts, export guardrails, and a versioned regression eval harness with deterministic scoring — the core path runs without an LLM API key.

PolicyGate

A runtime permission gateway for AI agent tool calls: the model proposes an action, deterministic code decides. Zod-validated policy engine returning ALLOW / APPROVAL / DENY, SOP-to-policy compilation, and an MCP-based tool-call integration.

codex-handoff

Verified session handoffs for long-running Codex CLI agents, rebuilding continuation state from git and test evidence instead of a chat summary. Installable CLI with CI and tests; only hands off at a safe turn boundary, with a bilingual README.

ContexCue

A cross-app memory and context layer for personal AI agents that gates what gets remembered instead of storing everything. MemoryGate scopes every record by person, task, evidence, privacy level, and TTL before it can reach a reply.

Mini-GPT

A from-scratch, decoder-only Transformer (nanoGPT-style) trained on character-level text. Full training loop, causal self-attention, and autoregressive sampling implemented directly in PyTorch.

What I Work With

Python · TypeScript · JavaScript · React · Next.js · FastAPI · SQL · LLM APIs · RAG · Agent Systems · MCP · Git

Currently

  • Master of Computing in Applied AI, NTU — Singapore
  • Focused on applied AI / agent engineering: data agents, RAG, reliable AI application design
  • Open to Applied AI / AI Engineering opportunities

Links

Pinned Loading

  1. codex-handoff codex-handoff Public

    Non-blocking compaction reminders and explicit, evidence-backed handoffs for long Codex sessions.

    Python 3

  2. CA6123-frontier-lit-agent CA6123-frontier-lit-agent Public

    Multi-agent literature synthesis system built with Google ADK, Gemini, and RAG. Group project for CA6123 (Agentic AI and Applications).

    Python

  3. Mini-GPT Mini-GPT Public

    This project is a hands-on implementation of a decoder-only Transformer language model, inspired by Andrej Karpathy's nanoGPT and Stanford's CS336. The goal is to clearly demonstrate the core compo…

    Python

  4. Data-Agent-Sandbox Data-Agent-Sandbox Public

    TypeScript