STATUS: OPEN FILED 07.23.2026 · JURISDICTION: AI/ML SYSTEMS & BLOCKCHAIN

GRAYSON LIM

AI/ML DEVELOPER — LLMS, RAG & AGENTIC SYSTEMS

I don't trust a system just because it runs. Before shipping anything — a RAG pipeline, an agent, a quiz generator — I want to have taken it apart and rebuilt the parts that don't hold up under real questions.

05 CASE FILES CLOSED 02 ASSIGNMENTS SERVED ACTIVE SINCE 2017
OPEN THE FILE
EVIDENCE 02

The Suspect Profile

Grayson Lim
HEIGHT OF STACKAI/ML + WEB3
BASE OF OPSSINGAPORE
ACTIVE SINCE2017

Recent Computer Science graduate who's spent the last year building at the intersection of AI, retrieval systems, and real production environments — now looking for a full-time role as an AI/ML Developer. During an internship as an AI Developer at Huiyi Intelligent Technology (中国惠医智能科技有限公司), I worked across an AI-powered contract review system built on multi-agent workflows, an autonomous social media agent, and text-to-video pipelines for client demos — hands-on with LLMs, RAG, and vector databases throughout.

For my capstone, I built SynapseQ, a full-stack Auto Quiz Generation system (FastAPI + Vue 3) that ingests lecture materials, builds a searchable topic hierarchy, and generates grounded MCQ quizzes using hybrid retrieval (Qdrant + BM25) and a multi-stage LLM pipeline — turning messy lecture content into structured, trustworthy quizzes.

Curious by nature, I learn fast and genuinely enjoy figuring out how things work and why. Outside of AI/ML, I've been exploring blockchain on the side. Now looking for a team building applied AI systems — RAG, agents, LLM infrastructure — that ships to real users.

KNOWN ASSOCIATE: AGENTIC KNOWN ASSOCIATE: QDRANT KNOWN ASSOCIATE: SOLIDITY
EVIDENCE 03

The Corkboard

EVERY PIN BELOW CONNECTS TO AT LEAST ONE CASE FILE FURTHER DOWN THE BOARD.

AI / ML

  • LLMs & Prompt Engineering
  • Retrieval-Augmented Generation (RAG)
  • Hybrid Retrieval (BM25 + Vector Search)
  • Vector Databases (Qdrant, PGVector)
  • Multi-Agent Systems / Agentic Workflows
  • LangChain / LangChain4j
  • Anthropic Claude API / OpenAI API

Backend & Full-Stack

  • FastAPI, Spring Boot, Node.js
  • Next.js, Vue 3
  • tRPC, Prisma
  • PostgreSQL, Redis
  • Docker
  • REST API Design

Blockchain

  • Solidity
  • Ethereum / EVM (Base L2)
  • Smart Contract Development
  • ethers.js / Web3.js
  • Uniswap V2 Integration

Automation & Integrations

  • Telegram Bot Development
  • Payment Integration (PayNow, HitPay)
  • Third-party API Integration (Twitter/X, Google Calendar)
  • Task Scheduling / Cron-based Systems
  • Real-time Systems (Pusher / WebSockets)

Product & Systems Thinking

  • System Architecture & Database Design
  • Bilingual / Internationalization Design
  • Human-in-the-loop / Semi-automated Workflow Design
EVIDENCE 04

The Case Files

CASE A · FILED 07.2026

The Contactless Handoff

CLOSED

PROBLEM

Small F&B vendors wanted a QR ordering flow that didn't hand a cut of every sale to a payment processor, but still needed live order tracking and a vendor console that didn't feel like an afterthought.

INVESTIGATION

Mapped the flow around a fee-free path first — scan, cart, client-generated PayNow QR — then layered a real-time channel on top so both sides see the same order state without polling.

SOLUTION

Built a QR ordering & payment PWA with a Pusher-powered live queue for the vendor console (orders, menu, categories), bilingual EN/中文 support, and an optional HitPay integration for merchants who want automated reconciliation and card payments.

FEE-FREE PAYNOW BY DEFAULT REAL-TIME VENDOR QUEUE NEXT.JS · TRPC · PRISMA · PUSHER
CASE B · FILED 04.2026

The Grounded Quiz

CLOSED

PROBLEM

Lecture materials pile up as unstructured slides and notes; turning them into quizzes that actually test the material — not just plausible-sounding questions — usually means manual work or an LLM that quietly hallucinates.

INVESTIGATION

Retrieval on raw lecture text showed the usual thin-RAG failure mode: keyword search missed paraphrased concepts, and pure vector search surfaced fluent but ungrounded matches.

SOLUTION

Built SynapseQ, a full-stack platform (FastAPI + Vue 3) that ingests lecture materials, builds a searchable topic hierarchy, and generates grounded MCQ quizzes through hybrid retrieval (Qdrant + BM25) feeding a multi-stage LLM generation pipeline.

HYBRID RETRIEVAL: QDRANT + BM25 MULTI-STAGE LLM GENERATION FASTAPI · VUE 3 · QDRANT
CASE C · FILED 11.2025

The Autonomous Correspondent

CLOSED

PROBLEM

Keeping a social presence on-voice and current means someone has to watch AI news, decide what's worth posting, write it up, and hit publish — every day, without fail.

INVESTIGATION

Broke the job into independent stages rather than one monolithic script, so each part could be tested, swapped, or rerun on its own without touching the rest of the pipeline.

SOLUTION

Built an agent-skills-based automation pipeline (Node.js + Claude API + OpenClaw) with four independent SKILL.md-defined modules — fetch, format, queue, post — that let the agent autonomously discover AI news, draft on-voice tweets, and schedule posts to Twitter/X.

4 INDEPENDENT SKILL MODULES NODE.JS · CLAUDE API · TWITTER/X API
CASE D · FILED 03.2024

The Instant Deploy

CLOSED

PROBLEM

Deploying an ERC-20 token contract normally means a dev environment, hand-edited constructor parameters, and a deploy script — too much friction for testing configurations quickly.

INVESTIGATION

The friction wasn't the Solidity itself — it was the round-trip between "I want to change one parameter" and "the contract is live on-chain."

SOLUTIONBuilt a Telegram bot that deploys ERC-20 token contracts with different configurations on request, collapsing the edit-compile-deploy loop into a single chat command.

CHAT-COMMAND CONTRACT DEPLOYS SOLIDITY · ETHERS.JS · TELEGRAM BOT API
CASE E · FILED 08.2023

The Early Tip-Off

CLOSED

PROBLEM

On Friend.tech, being early to a new influencer's key sale mattered more than almost anything else — and manually watching for new joins didn't scale past a handful of accounts.

INVESTIGATION

Needed a way to detect new joins and flag influencer accounts fast enough to act before the price moved, without babysitting a dashboard.

SOLUTIONBuilt a bot that detects new joins and influencer accounts on Friend.tech in real time and auto-buys according to configurable rules — budget caps, take-profit/stop-loss thresholds, and more.

REAL-TIME NEW-JOIN DETECTION CONFIGURABLE TP/SL RULES NODE.JS · ETHERS.JS · WEB3.JS
EVIDENCE 05

Assignment Log

SEP 2025 — APR 2026

AI Developer, Huiyi Intelligent Technology

AI team building enterprise LLM, RAG, and multi-agent solutions for B2B SaaS clients — 中国惠医智能科技有限公司, Jiangsu, China.

  • Designed backend architecture and database schema (LangChain4j + Spring Boot) for an AI-powered contract review system, including system diagrams and ER design for the Audit Rule Module.
  • Built and deployed OpenClaw, an autonomous agent for end-to-end social media automation — Skills-Markdown tool definitions, shell scripting, and the Twitter API to automate extraction, formatting, and publishing.
  • Evaluated and deployed text-to-video generation pipelines (local & cloud) with character/scenario customization, producing client demo content and cutting production cost and turnaround time.
  • Worked hands-on with LLMs, RAG, and vector databases (PGVector, Qdrant), Docker, and Redis in a production B2B SaaS environment — recognized by the General Manager for strong technical foundation and clear cross-stakeholder communication.
SEP 2020 — JUN 2021

Junior Software Engineer, Ebuy Pte Ltd

Built and maintained e-menu ordering platforms for F&B clients including ShangPin Hotpot, Hippot Hotpot, and Upin Hotpot — Singapore.

  • Designed backend architecture and database schemas using Node.js, Express, MongoDB, and Mongoose, deployed with PM2 and Nginx.
  • Built real-time features using WebSockets to support live order tracking and updates.
  • Worked cross-functionally with design and product teams to iterate on UI/UX for each client's e-menu system.
  • Maintained multiple concurrent client deployments, balancing feature development with production reliability — recognized by the IT Head of Department for consistently high productivity and quality of output.
EVIDENCE 06

The Training Record

AUG 2023 — APR 2026

Singapore Institute of Technology

Bachelor of Science with Honours in Computing Science — Achieved Honours with Merit.

APR 2018 — MAR 2021

Singapore Polytechnic

Diploma in Information Technology — GPA 3.614 / 4.0.

COMMENDATIONS ON FILE

Edusave Certificate of Academic Achievement MAR 2021 · MAY 2018 · DEC 2017
Edusave Merit Bursary MAR 2021 · MAR 2018 · DEC 2017
Director's List — Meritorious academic performance (top 10%) SEP 2016
EVIDENCE 08

Open a Direct Line

No forms, no runaround. Pick a channel and the line opens straight to me.

EMAIL graysonlimwork@gmail.com LINKEDIN /in/graysonlck GITHUB /greyspace-eth FULL FILE (RESUME) DOWNLOAD PDF ↓