Ridhwan Amin

// about

About me

Ridhwan Amin, AI and ML researcher, portrait

MEXT Scholar 2022

IEEE Member

UTP Researcher

M.Sc. EEE

PETRONAS Digital Alumni

I'm Ridhwan Amin, an AI/ML researcher at Universiti Teknologi PETRONAS (UTP) in Malaysia. My research sits at the intersection of deep learning and geophysical signal processing, specifically using implicit neural representations and generative priors to reconstruct missing seismic data.

Beyond research, I consult with teams that need production-ready ML infrastructure: RAG pipelines that actually retrieve the right things, anomaly detection systems that don't cry wolf, and MLOps setups that let your team iterate without fear.

I received a MEXT scholarship from the Japanese government and am a member of IEEE. I also write about ML, research craft, and working in energy-domain AI on my Substack.

Get in touch View services

// capabilities

What I build

systemprompttemperaturecontext windowfine-tuneRLHFtokenizerbeam searchtop-pchain-of-thoughtsystemprompttemperaturecontext windowfine-tuneRLHFtokenizerbeam searchtop-pchain-of-thought

⟨input⟩→attention→embedding→logit→sample→⟨output⟩→⟨input⟩→attention→embedding→logit→sample→⟨output⟩→

GPT-4oClaude 3Llama 3MistralGeminiLoRAQLoRARAGORPOSFTGPT-4oClaude 3Llama 3MistralGeminiLoRAQLoRARAGORPOSFT

LLM Engineering

Designing, fine-tuning, and deploying large language models for real-world tasks. From prompt engineering and RAG pipelines to RLHF and LoRA fine-tuning on domain-specific corpora.

GPT-4o Claude Llama 3 LangChain RLHF LoRA

Generative AI

Building generative systems: diffusion models, VAEs, and multimodal pipelines. Research-grade experimentation combined with production-feasible architectures.

Diffusion Stable Diffusion VAE ControlNet Fine-tuning

import torchfrom transformers import AutoModel

class RAGPipeline: def __init__(self, model_id): self.model = AutoModel .from_pretrained(model_id)

def encode(self, text): tokens = self.tokenizer(text) with torch.no_grad(): return self.model(**tokens)

def hybrid_search(query, k=5): dense = faiss_index.search(query, k) sparse = bm25.get_top_n(query, k) return rerank(dense + sparse)

async def stream_response(prompt: str): async for chunk in llm.astream(prompt): yield chunk.contentimport torchfrom transformers import AutoModel

class RAGPipeline: def __init__(self, model_id): self.model = AutoModel .from_pretrained(model_id)

def encode(self, text): tokens = self.tokenizer(text) with torch.no_grad(): return self.model(**tokens)

def hybrid_search(query, k=5): dense = faiss_index.search(query, k) sparse = bm25.get_top_n(query, k) return rerank(dense + sparse)

async def stream_response(prompt: str): async for chunk in llm.astream(prompt): yield chunk.content

Python Development

Fluent in the full Python ML stack. From async FastAPI services to GPU-accelerated PyTorch training loops, with clean, testable, production-ready code.

PyTorch FastAPI NumPy Pandas scikit-learn asyncio

Cloud Development

Deploying and scaling ML workloads on AWS and Azure. Containerised inference, managed vector stores, serverless pipelines, and IaC with Terraform.

AWS Azure Docker Kubernetes Terraform CI/CD

// projects

Featured work

Hospital Bed Explorer homepage showing a map and real-time bed availability across Malaysia

Featured

Malaysia Hospital Bed Explorer

Real-time dashboard surfacing available hospital beds across every Malaysian state, built on Cloudflare Workers and the Ministry of Health open-data API.

Cloudflare Workers React MOH Open Data REST API

Stethoz homepage showing the professional network for healthcare workers with a countdown to launch

Co-builder

Stethoz: Professional Network for Healthcare

Co-built a LinkedIn-style networking platform for doctors, nurses, and allied health professionals to share clinical knowledge, discuss cases, and discover locum opportunities.

Healthcare Tech Social Platform Malaysia Made in MY 🇲🇾

Co-builder

REDO AI Solutions

AI consultation and cloud training platform helping organisations accelerate digital transformation through hands-on programmes and expert guidance.

AI Education Cloud Training Consultation

Azure Hybrid RAG Pipeline

Hybrid retrieval-augmented generation on Azure: combines dense vector search with BM25 keyword retrieval and an LLM reranker for high-precision enterprise document Q&A.

Azure AI Search LangChain FAISS OpenAI FastAPI

// blog

Writing

All posts

// faq

Frequently asked

What does a typical engagement look like?

Six to twelve weeks, embedded part-time with your team. I scope a concrete deliverable up front (a deployed RAG service with evals, an agent pipeline with observability, a training and serving stack with monitoring), then ship it. Open-ended retainers are possible but I prefer shape before commitment.

How do you measure "success" on an AI project?

Evals first, vibes second. Before I touch a prompt or a model, I write a small labelled set and an eval harness that runs on every change. If I can't measure it, I won't pretend to be improving it.

What stack do you reach for?

Python + FastAPI on the backend. pgvector or Qdrant for retrieval. Modal or Cloudflare Workers for serving, depending on latency budget. PyTorch when I'm training, W&B for tracking. TypeScript + Astro on the web. I'm stack-pragmatic; I'll use what your team already runs if it gets the job done.

Do you handle deployment and ongoing serving?

Yes. ML that doesn't reach users isn't really ML. I set up the CI, the canary rollout, the monitoring dashboards, the alerting. I'll hand the wheel back to your team with runbooks; I won't leave you a notebook and a prayer.

Available for full-time / contract / co-founding roles?

Contract-first, indefinitely. Open to interesting co-founding conversations if the problem is genuinely ML-shaped and the team has real domain context. Not looking for full-time employment right now.

// contact

Let's work together

Whether you need ML infrastructure, a research collaboration, or just want to talk about seismic imaging, I'm reachable.

View services and pricing