SWE @ DIS · CS @ NTU

Audric Yap

Builder across different domains, consistently using AI to compress time-to-output, explore solution space faster, and ship more.

AI-native

Builder

Iterative

Workflow-driven

Focus

"Turning LLMs into leverage."

[Work]

GradeSail

deployed

AI marking for schools and tuition centres: scanned scripts are transcribed and marked per rubric criterion, and teachers review and override every score before anything reaches a student.

#Next.js#tRPC#Cloudflare_Workers#Cloudflare_Queues#Prisma#Postgres#Clerk#AI_SDK

Applied AI + async processing + B2B SaaS

rook

research and integration

A No-Limit Hold'em poker AI in Rust and Python: depth-limited counterfactual-regret search at decision time, backed by a neural value network, run as a pre-registered research programme.

#Rust#PyTorch#CUDA#Python

Game-theoretic search + ML research programme

Streaming Data & Media Platform

deployed

Backend, media and platform engineering on an event-driven system that ingests, transforms and integrates high-volume heterogeneous data and serves it to operational users.

#Python#C##TypeScript#Kafka#MinIO#MediaMTX#Keycloak#OpenShift#Helm

Distributed systems + streaming media + platform engineering

AI Mafia

open source

LLM agents play a 17-role game of Mafia against each other, bluffing, fake-claiming roles and voting, with each agent seeing only what its role should know.

#TypeScript#Vercel_AI_SDK#Zod#Ink

Multi-agent simulation + game engine

VentureLane

prototype or internal

RAG-powered discovery over 4,300+ Y Combinator startups: hybrid semantic and keyword retrieval, and a chat assistant that grounds every answer in numbered citations to the records it retrieved.

#Next.js#Postgres#pgvector#OpenAI_embeddings#Docker

Retrieval-augmented search + assistant

[About]

Computer Science undergraduate at NTU and software engineer at the Digital & Intelligence Service, on the MINDEF/SAF Digital Specialist work-learn scheme: full-time engineering alternating with academic semesters.

I build with AI as a force multiplier across my entire workflow, not just in the final product. I use it to reason faster, explore alternatives, and compress iteration cycles, while keeping judgment and responsibility human. I also care about shipping usable tools and refining them under real use, not demos that only work in best-case conditions.

Beyond the featured work: a local-first communications archive that gives AI agents citable, paginated access to message history with no LLM in the retrieval path, and the occasional hackathon build under time pressure.

Base

Singapore

Stacks

Python, TypeScript, Rust, C#/.NET · Kafka, Kubernetes/OpenShift, Cloudflare · LLM APIs

[Contact]

Get in touch.

Open to discussions regarding AI-native products, high-leverage tooling, and hard technical problems.

Emailaudricyapzh@gmail.com

GITHUB LINKEDIN