APN®
AARON PATRICK NATHANIEL
0%

ENGINEERING THE INTERFACE

01 / PORTFOLIO Senior AI Engineer · Full-Stack Builder

AARON PATRICK NATHANIEL

Production AI.Built to perform.

Senior AI Engineer at Mindvalley building production-grade computer vision, generative AI and RAG systems, recommendation engines, MLOps, APIs, and full-stack SaaS products.
9+years in production AI
15Mimages used to fine-tune face recognition models
98%1:1 face verification accuracy
700+Chrome extension users
100+GitHub stars (Disp8ch)
15+shipped products (apps, SaaS, extensions, games)
SCROLL TO EXPLORE↓

From idea to production

Experience: 9 years of shipping

01

Mindvalley

Senior AI Engineer

Personalized recommendation systems across multiple digital products: candidate generation, ranking, content mix, evaluation, and safeguards to improve relevance, discovery, and long-term user outcomes.

02

Innov8tif Solutions

Senior AI Engineer & Full-Stack Developer

Started in computer vision for fraud detection and eKYC, then moved into LLMs, RAG systems, and MLOps. Led production LLM applications, enterprise RAG solutions, and automated model deployment infrastructure.

Eight production systems, with the results ↓

03

Education

Multimedia University

Bachelor in Computer Science (Data Science), Multimedia University, 2015 to 2018. CGPA 3.74

Built by hand.

Nine years of shipping, in one small figure.

Collectible figure of Aaron in a brown overshirt, holding a laptop with one foot on a football

SCROLL TO TURN · THEN FALL IN MOVE TO STEER · CLICK TO PULSE

THE DAY
JOB.

Nine years turning research into systems that run every day. Fraud detection, identity verification, document intelligence, enterprise RAG, and the pipelines that ship them.

FROM 2018
TO NOW.

SCROLL DOWN · EXPLORE RIGHT

  1. 01

    2015 to 2018

    Multimedia University

    Bachelor in Computer Science, Data Science.

    3.74CGPA

  2. 02

    2018

    Innov8tif, Full-Stack Engineer

    Computer vision for fraud detection and eKYC, in Python and OpenCV.

    2018Joined Innov8tif

  3. 03

    2019

    First patent filed

    Microprint verification for Malaysian identification documents. Granted in 2021 (MY184165A).

    Patent 1Malaysia

  4. 04

    2020

    Document authentication patent

    Authenticating identification documents with several algorithms and deep learning. Granted in 2022 (MY192715A), as first-named inventor.

    Patent 2Malaysia

  5. 05

    2021

    US patent filed

    Judging whether a security hologram is real from live video. Granted in the United States in 2024 (US12026932B2).

    US patentGranted 2024

  6. 06

    Innov8tif

    Our own face recognition model

    A year of cleaning data and training on 15 million images produced an in-house model that beat InsightFace and the paid solutions we benchmarked at the time. Took it through NIST's face recognition evaluations, where my cropping and data ideas drove the September 2026 gains.

    15MTraining images

  7. 07

    Innov8tif

    Face search in production

    Built on that model: a FAISS face search service that checks every new image against the archive.

    10K to 15KImages a day

  8. 08

    Innov8tif

    Vision on the edge

    Compressed production models to run fully offline on phones, and ran object detection in the browser with WebAssembly.

    <100 MBPer model

  9. 09

    Innov8tif

    Documents and OCR

    Rebuilt identity card text extraction, moving from a CTPN and CRNN pipeline to PPOCR, with NER on top for structured fields.

    4 s → 1.5 sInference per card

  10. 10

    Innov8tif

    LLMs and RAG

    Internal LLM and RAG applications for about 100 employees, with LoRA fine-tuning on the company handbook.

    ~90%Accuracy, ~60% fewer hallucination issues

  11. 11

    Innov8tif

    MLOps

    Introduced the company's first full MLOps pipeline with DVC, GitHub, Jenkins, and MLflow, and taught the team to use it.

    2.5 h → 20 minDeployment time

  12. 12

    Innov8tif

    Leading the team

    Led two junior engineers and two interns at the same time. Code reviews, onboarding, technical interviews, and the quarterly roadmap.

    4People

  13. 13

    2026 to now

    Mindvalley, Senior AI Engineer

    Personalised recommendations across multiple digital products.

    NowRecommendations

SYSTEMS
IN PRODUCTION.

Client names stay private. The engineering does not.

Identity document authentication: Phone video or images, then Classify document, then Microprint, ghost image, hologram vs glare, then Combine the checks, then Genuine or reject Phone video orimages Classifydocument Microprint,ghost image,hologram vsglare Combine thechecks Genuine orreject
Problem
Onboarding has to tell a genuine identity card from a printed or edited copy, from a phone camera, in seconds.
What I built
The authentication methods behind three granted patents: microprint checked on contrast-enhanced images; a pipeline that classifies the document, compares the ghost image with the photo, checks colour capture, and finds the security hologram; and a hologram check from live video that uses where the hologram sits relative to the glare to reject printed copies.
Decisions
  • Combine several independent checks, so one forgery trick is never enough.
Stack
  • Python
  • OpenCV
  • Deep learning classifiers
My role
Co-inventor on all three. See the patents ↓

Face recognition model: 15M cleaned images, then Crop and align (new cropping), then Train, then Internal benchmark vs vendors, then NIST submission, then Production 15M cleanedimages Crop andalign (newcropping) Train Internalbenchmarkvs vendors NISTsubmission Production Iterate on data and crops
Problem
Off-the-shelf and paid face recognition was not accurate enough on the faces and capture conditions the product actually saw.
What I built
An in-house face recognition model. About a year of cleaning data and training on 15 million images, reaching about 98% 1:1 accuracy on our internal benchmark. On that benchmark, at the time, it beat InsightFace and the paid solutions we evaluated, with far less training data than the big vendors.
NIST 1:N
Tested in NIST's 1:N identification evaluation, alongside the largest face recognition vendors in the world. In mugshot searches across a 1.6 million face gallery, the right person came back as the top match 91% of the time.
What came next
My research direction, a new face cropping approach and more mugshot-style training data, drove the gains in the company's September 2026 NIST submission: roughly 30 to 40% fewer missed matches on border and visa photos.
Decisions
  • Invest in data cleaning before reaching for new architectures.
  • Own the model instead of paying per call for a vendor API.
Stack
  • Python
  • PyTorch
  • OpenCV
My role
Led the research, from data cleaning to training.

Face search at scale: Image in, then Detect and align, then Face embedding, then FAISS index, then Top-k candidates, then Threshold, then Review or alert Daily new images Image in Detect and align Face embedding FAISS index Top-k candidates Threshold Review or alert
Problem
Fraud teams need to know whether a new applicant's face has been seen before, across a growing archive, every day.
What I built
A production face search service on FAISS, built on the in-house recognition model for its embeddings.
Decisions
  • A FAISS vector index, so a growing archive stays searchable every day.
  • Thresholds tuned against the review workload.
Stack
  • Python
  • PyTorch
  • OpenCV
  • FAISS
  • FastAPI
  • Docker
My role
Built the service.

Vision on the edge: Camera, then On-device model under 100 MB, then Liveness or detection, then Result OFFLINE, NO NETWORK Camera On-device model under100 MB Liveness or detection Result
Problem
Mobile and browser checks had to run fully on the device, with no connection.
What I built
Production vision models compressed to under 100 MB each for fully offline mobile inference, face monitoring tools for fraud detection, and object detection in the browser with WebAssembly.
Decisions
  • Compress the models to fit on a phone instead of calling a server.
  • Run detection in the browser through WebAssembly, so nothing has to be uploaded.
Stack
  • Python
  • OpenCV
  • WebAssembly
  • On-device inference
My role
Built the face monitoring tools and the WebAssembly detection.

Document intelligence: Card image, then Classify document, then Detect and read text (CTPN + CRNN, then PPOCR), then NER, then Structured fields Card image Classifydocument Detect and readtext (CTPN +CRNN, thenPPOCR) NER Structuredfields
Problem
Pulling personal details off a Malaysian identity card (MyKad) had to be fast enough for live onboarding.
What I built
Replaced a CTPN text detector and CRNN recogniser pipeline with PPOCR, cutting inference from about 4 seconds to about 1.5 per card. Also image classification for document types, and bank statement processing.
Decisions
  • Add NER after OCR, so the output is structured fields instead of raw text.
Stack
  • PPOCR
  • CRNN
  • CTPN
  • NER
  • Python
My role
Rebuilt the text extraction pipeline.

Enterprise LLM and RAG: Documents, then Chunk, then Embed, then Vector store, then Retrieve, then LLM, local or API, then Guardrails, then Answer Documents Chunk Embed Vector store Retrieve LLM, local or API Guardrails Answer Evaluate, then tune chunking and retrieval
Problem
Employees needed reliable answers from the company handbook and documents.
What I built
Internal LLM and RAG applications on open-source models: a handbook assistant, and a bank statement and document classification application. LoRA fine-tuning on the handbook, plus chunking, retrieval, and parameter tuning. About 90% accuracy and about 60% fewer hallucination and guardrail issues, and the custom LoRA models matched leading provider performance for the business use case.
Decisions
  • Measured local LLM deployment against the GPT API on cost and performance before choosing.
  • Hallucination testing and guardrails built in, not added afterwards.
Stack
  • Python
  • Ollama
  • LoRA
  • FAISS
  • Streamlit
My role
Led development of the production LLM applications.
Next
Decant, below, is the tool I wished I had while building this.

MLOps platform: Data prep change, then DVC and GitHub commit, then Jenkins, then Train, then MLflow registry, then Docker, then FastAPI, then Deploy Data prep change DVC and GitHub commit Jenkins Train MLflow registry Docker FastAPI Deploy
Problem
Training and deployment were manual and slow, and they depended on one person's machine.
What I built
DVC for data versions, GitHub for code, Jenkins for automation, and MLflow for experiments and the model registry. Docker optimised and services moved to FastAPI for faster inference.
Decisions
  • Every model traces back to the exact data and code that made it, so anyone can retrain or roll back.
Stack
  • DVC
  • GitHub
  • Jenkins
  • MLflow
  • Docker
  • FastAPI
My role
Introduced the company's first full MLOps pipeline.

Recommendations: Catalogue and user signals, then Candidate generation, then Ranking, then Content mix, then Safeguards, then Feed Catalogueand usersignals Candidategeneration Ranking Content mix Safeguards Feed Evaluation metrics feed the next ranking
Problem
Help people find the right next thing across several digital products, and keep doing it well over the long term.
What I built
Personalised recommendation systems: candidate generation, ranking, content mix, evaluation metrics, and safeguards for relevance, discovery, and long-term user outcomes. In progress, so no public metrics yet.
Decisions
  • Evaluation metrics and safeguards are part of the pipeline from the start.
My role
Senior AI Engineer on recommendations.

3 GRANTED
PATENTS.

  • US12026932B2

    Method to determine authenticity of security hologram

    Filed 2021. Granted 2 July 2024. United States.

    Co-inventor

    VIEW ON GOOGLE PATENTS ↗
  • MY192715A

    Method for authenticating identification documents

    Filed 2020. Granted 2022. Malaysia.

    First-named inventor

    VIEW ON GOOGLE PATENTS ↗
  • MY184165A

    A method to verify authenticity of a Malaysian identification document

    Filed 2019. Granted 2021. Malaysia.

    Co-inventor

    VIEW ON GOOGLE PATENTS ↗
The granite ridge leading toward Low's Peak at Mount Kinabalu
08 / KINABALU TO NOTE

The mountain on the note is the one I climbed.

Low's Peak, Mount Kinabalu, 4,095 m.

Reverse of a Malaysian RM100 note featuring the Mount Kinabalu engraving

DECANT.

Teach it. Give it knowledge. Give it abilities. Prove it works.

  1. 01Teach
  2. 02Knowledge
  3. 03Abilities
  4. 04Prove

The question

I built Decant to turn a small open model into a specialist on your own documents: synthetic training data, LoRA fine-tuning, retrieval, tools, guardrails, and a held-out final exam. Then I tried to prove it worked.

Manuals
8
Student model
Qwen3-4B
Exam questions
285
Experiments
26+
Decant home screen: Teach any model something new. Decant goal step: describe what the model should learn. Decant guardrails step: topics the model must decline. Decant knowledge step: manuals indexed for retrieval. Decant dataset step: a quality-checked training set.

01 / Teach any model something new.

Round 2 · 285 questions · independent judge

Retrieval did the work. Fine-tuning did not.

Untuned, no retrieval 18.8%
Fine-tuned + retrieval 83.8%
Untuned + retrieval 91.2%
Untuned + cross-encoder reranker 93.8%
Factual accuracy on 8 plant-safety manuals. Sampling noise alone moves a score by up to 6.2 points, so smaller gaps are never ranked.

Round 2 · general ability

Fine-tuning made it forget. One mix fixed it.

Base model 88.2%
After fine-tuning 35.3%
Fine-tuning + 20% general data 91.2%
Mixing in about 20% general instruction data restored base-model ability. The forgetting came from narrow training data, not from the model being small.

Inside retrieval

716 chunks. One question.

  1. 01Embed the question and find its nearest chunks.
  2. 02A cross-encoder reranker reorders them. It was the single best change.
  3. 03Fit the best passages into the 1800 token budget.
  4. 04One wrong passage in the mix, and the model loses the answer.

Real chunk layout of the study corpus. The query path is an illustration.

Negative results I kept

Two things that did not work.

  • 95.5 → 91.0

    Hybrid retrieval hurt

    Dense plus BM25 scored lower than dense alone.

    Round 1
  • 85.0% vs 78.7%

    Three examples beat the adapter

    Few-shot prompting out-scored the fine-tuned model.

    Round 1

How it was measured

Honest numbers, or none.

  • Report the noise floor, and never rank gaps smaller than it
  • Two scorers on every answer, shown side by side
  • Measure under the budget you ship with

Retrieval is how the model knows things. Fine-tuning is how it writes.

READ THE
FULL STUDY ↗

HOW I
LEAD.

I led a team of four at Innov8tif: two junior engineers and two interns. Fast models were only half the job. The other half was making sure the team could run, fix, and retrain them without me in the room.

  1. 01

    One pipeline

    OptimisedChange the data prep, commit, and training runs on its own to a tracked, deployable model.

    MaintainableI taught the whole team the workflow, so training no longer depended on one person's machine.

  2. 02

    Code reviews

    OptimisedEvery change reviewed for speed: smaller models, faster inference.

    MaintainableAnd for whether the next person can read, test, and change it.

  3. 03

    Onboarding and hiring

    OptimisedPrepared the technical interviews, to hire for the work.

    MaintainableNew engineers and interns learned the same pipeline and standards as everyone else, so nothing depended on one person.

  4. 04

    Quarterly roadmap

    OptimisedPlanned each quarter with the team.

    MaintainablePeople built what they had helped plan, and owned it after release.

The test of a good system is the second person who has to maintain it.

SKILLS.
END TO END.

Python · TypeScript · JavaScript · C++ · SQL · PyTorch · TensorFlow · Hugging Face · OpenCV · YOLO · PPOCR · NER · Scikit-learn ·
OpenAI · Gemini · Ollama · LangChain · RAG · LoRA fine-tuning · FAISS · Guardrails · Prompt evaluation ·
FastAPI · React · Next.js · Docker · Kubernetes · Airflow · MLflow · DVC · GitHub Actions · Jenkins · GitLab CI/CD · AWS · GCP · Azure · Playwright · Vitest · PyTest · Stripe ·
11B / DELIVERY
WHAT I CAN OWN

END TO
END.

Move to complete the system.
Scroll to follow the work.

  1. 01 / DEFINE

    requirements and scoping,

  2. 02 / DESIGN

    data prep and knowledge-base design, model and provider evaluation,

  3. 03 / BUILD

    RAG design, retrieval tuning and fine-tuning, hallucination testing and guardrails,

  4. 04 / SHIP

    backend APIs, auth and frontend, testing, Docker, CI/CD and MLOps,

  5. 05 / OPERATE

    then monitoring, feedback loops, and production support.

Hiring a Senior AI Engineer
or an AI lead?

DISP8CH / LOCAL-FIRST AI
Disp8ch local-first AI workspace hero screen
14 / THE BIG MOMENT

DISP8CH

Local-first AI workspace for real work.

Turns plain-English requests into complete workflows, decisions, and tasks that execute on your machine with approval gates for risky actions.

  • Agentic chat
  • Visual workflow automation
  • Multi-agent "Council" deliberation
  • Supervised delivery
  • Independent review
  • Verifiable outcomes

Works with fully local models (Ollama, LM Studio, llama.cpp) or cloud providers, with no API keys required for core functionality. A better Hermes and OpenClaw alternative.

VISIT DISP8CH ↗

SAAS
PRODUCTS.

All SaaS built solo, full-stack: auth, database, dashboards, Stripe payments, backend APIs, cloud deployment. Around 200 monthly users across products.

01 / 04
RenderCrib AI hero screen
50+ STYLESPRECISE REGION EDITINGPRESERVES LIGHTING, MATERIALS, AND ARCHITECTURE

RenderCrib AI

AI interior design for teams who demand polish. Room photos become photoreal design concepts across 50+ styles, with precise region editing that preserves lighting, materials, and architecture.

OPEN PRODUCT ↗
02 / 04
Picscape AI hero screen
PHOTO RESTORATIONVIRTUAL TRY-ONOBJECT CLEANUP

Picscape AI

Turn ordinary photographs into professional photoshoots. Browser studio for photo restoration, background removal, portrait stylization, virtual try-on, and object cleanup.

OPEN PRODUCT ↗
03 / 04
promptPDF hero screen
MERGESPLITCONVERTEXTRACTSUMMARIZE

promptPDF

Transform PDFs with natural language. Describe what you need, whether merge, split, convert, extract, or summarize, and AI does it in the browser.

OPEN PRODUCT ↗
04 / 04
EditToBase64 hero screen
100+ BROWSER TOOLS100% CLIENT-SIDENOTHING UPLOADED

EditToBase64

100+ browser tools for converting, encoding, and processing files: images, PDFs, data, and developer utilities, all 100% client-side with nothing uploaded.

OPEN PRODUCT ↗

IOS APPS
& GAMES.

His Words App Store page
01 / APP STORE

His Words, AI Bible Study

AI-powered Bible study companion: instant verse explanations, generated study notes, historical context, audio narration, reading plans, and multiple translations.

Censorie App Store page
02 / APP STORE

Censorie

Privacy-first photo censoring. On-device face detection and text recognition; blur, pixelate, black-bar, or face-replace before sharing. Strips EXIF on export.

NailScan AI App Store page
03 / APP STORE

NailScan AI

AI nail health and style. Snap a photo for instant texture, color, and health assessment, virtual style try-ons, care routines, and an AI care assistant.

Word Step App Store page
04 / APP STORE

Word Step (game)

Word Ladders and Sea Battles. 500 hand-crafted levels across five puzzle types, Word Wheel bonus rounds, and treasure-hunting sea battles, in a hand-painted aesthetic.

04 / COMPLETE COLLECTION

Ideas become products.

Research, design, engineering, testing, launch, and support across four independent App Store releases.

END TO END / SHIPPED SOLO

SCROLL DOWN / EXPLORE RIGHT

4 WORKING PRODUCTS.
7 DAYS.

4 working GPT-5.6 products in one week. Proof of fast, high-quality demo delivery.

DemoForge Build Week project cover
01

DemoForge

turns recordings, screenshots, code, and natural voice into an editable, motion-designed product demo

EchoWorld Build Week project cover
02

EchoWorld

turns a place photo or room idea into a spatial world you can step inside

ManualZero Build Week project cover
03

ManualZero

turns a few product photos into clear, safety-first guides and offline manuals

TriWear Build Week project cover
04

TriWear

evidence-backed product comparisons, virtual try-on, and used-item checks

LET'S SHIP IT.

Open to freelance and project-based work, remote or after-hours. If you need someone who can understand the problem, design the solution, and ship it to production, let's talk. Also open to remote Senior AI Engineer and AI lead roles. The short version is on the hiring page.