Building production-grade AI systems

Hi, I'm Sameet Asadullah

AI/ML Engineer

AI/ML engineer with 4+ years building LLM, RAG and computer-vision systems that scale — from vector search and model serving to MLOps pipelines supporting up to 20M daily requests.

View my workGet in touch
SA
Sameet Asadullah
0+
Years of experience
0M+
Daily requests served
0+
ML projects shipped
A quick introduction

About Me

SA
Sameet Asadullah

Sameet Asadullah

AI/ML Engineer

EducationM.AI/ML

I'm an AI/ML engineer building and deploying production systems end to end — from LLM-powered features and RAG pipelines to vector search, model serving and observability. I work hands-on across the full lifecycle, using Python, FastAPI, Docker, Kubernetes and Triton on AWS and GCP.

I hold a Master's in Artificial Intelligence and Machine Learning from the University of Adelaide and a Bachelor's in Computer Science from FAST-NUCES. I bring a practical, product-minded approach to engineering with a strong focus on performance and reliability.

What I do

LLMs & RAG
Semantic Search
Generative AI
Computer Vision
MLOps & Serving
Backend APIs

Education

Master of Artificial Intelligence and Machine Learning

The University of Adelaide

Download résumé
Where I've worked

Experience

  • AI Engineer

    Jul 2024 — Jan 2026

    Stealth Startup · Remote

    • Designed and rolled out a vector-based search engine with Elasticsearch and transformer embeddings, improving product search relevance by around 35–45%.
    • Integrated an LLM-driven recipe feature powered by RAG, matching required ingredients to supermarket products and adding them straight to the cart.
    • Built web scrapers for Coles, Aldi, IGA and Woolworths that keep thousands of product listings up to date in real time.
    • Created a product normalization pipeline that cleaned up messy product names and generated embeddings, improving both search ranking and product grouping.
    ElasticsearchTransformersRAGPython

  • Machine Learning Engineer

    Feb 2025 — Jul 2025

    Add Life Technologies · On-site

    • Optimized a real-time body tracking application by integrating MediaPipe with Unity and tuning performance for mobile, cutting processing latency by 40% on mid-range devices.
    • Led end-to-end development of a cross-platform AI-powered mobile app — backend APIs in FastAPI, front end in Flutter — for seamless real-time video stream handling.
    • Improved system reliability and scalability through stress testing, memory leak fixes and asynchronous data handling, stabilizing performance under 10+ concurrent video streams.
    MediaPipeUnityFastAPIFlutter

  • Machine Learning Engineer

    Aug 2022 — Jan 2024

    Vyro · Remote

    • Built a bespoke serving architecture for ImagineArt, the second most popular AI art generator in the US, using FastAPI and Docker — supporting 30+ ML models at up to 20M requests/day with 99.5% uptime on Kubernetes.
    • Deployed ML models for Phototune (10M+ downloads) on Triton Inference Server, optimizing response time to 2–3 seconds.
    • Implemented a Docker-based serverless architecture for AvatarMe using Runpod and AWS (ECS, S3), cutting avatar-training turnaround from hours to 15 minutes.
    • Designed GitHub Actions CI pipelines achieving 97% test coverage, reducing production errors by 95%.
    • Engineered a Python SDK to host any Stable Diffusion workflow in production, cutting feature-hosting time by 80% through reuse.
    FastAPIDockerKubernetesTriton Inference ServerStable Diffusion
Some things I've built

Featured Projects

VStealth Startup

Vector-Based Product Search

Vector-based product search engine built with Elasticsearch and transformer embeddings, improving search relevance by 35–45% across the catalog.

ElasticsearchTransformersPython
Company project
RStealth Startup

RAG Recipe Copilot

LLM-driven recipe feature powered by RAG — enter a dish name and get a recipe with ingredients semantically matched to supermarket products and added straight to the cart.

RAGPythonElasticsearch
Company project
MStealth Startup

Multi-Retailer Web Scrapers

Real-time web scrapers for Coles, Aldi, IGA and Woolworths that keep thousands of product listings fresh and complete.

PythonREST APIs
Company project
PStealth Startup

Product Normalization Pipeline

Product normalization pipeline that cleans messy product names and generates embeddings to improve search ranking and product grouping.

PythonTransformers
Company project
RAdd Life Technologies

Real-Time Body Tracking

Real-time body tracking app integrating MediaPipe with Unity, tuned for mobile deployment with a 40% latency reduction.

MediaPipeUnity
Company project
CAdd Life Technologies

Cross-Platform AI Mobile App

Cross-platform AI-powered mobile app with a FastAPI backend and Flutter front end for seamless real-time video streaming.

FastAPIFlutter
Company project
IVyro

ImagineArt Model Serving

Bespoke ML serving architecture for ImagineArt — 30+ models, up to 20M requests/day and 99.5% uptime on FastAPI, Docker and Kubernetes.

FastAPIDockerKubernetes
Company project
PVyro

Phototune on Triton

Triton Inference Server deployment for Phototune (10M+ downloads), optimized for a 2–3 second response time.

Triton Inference ServerDocker
Company project
AVyro

AvatarMe Serverless Training

Docker-based serverless architecture for on-demand avatar training using Runpod and AWS, cutting turnaround from hours to 15 minutes.

DockerAWS
Company project
SVyro

Stable Diffusion Python SDK

Python SDK for hosting any Stable Diffusion workflow in production, cutting feature-hosting time by 80% through reuse.

Stable DiffusionPython
Company project
A

ApplyGraph

Session-based agentic AI job copilot using FastAPI, LangGraph and PostgreSQL + pgvector to analyze job fit, tailor resumes, draft outreach and persist semantic memory across chat threads.

FastAPILangGraphPostgreSQL
View code
M

MergeWise

AI-powered pull request reviewer combining RAG with FAISS-based context retrieval, LLM reasoning and GitHub Checks — inline or via a Celery/Redis queue.

RAGFAISSFastAPI
View code
A

AutomateIt

Home automation system using a CNN and React Native, recognizing Urdu voice commands with over 85% accuracy to control household appliances.

CNNReact NativeMongoDB
View code
T

Temporal Financial Forecasting

Deep learning pipeline for multi-horizon financial time-series forecasting with RNNs, GRUs and LSTMs — best GRU model reaches RMSE ≈ 0.024.

PyTorchPython
View code
C

CNN Benchmark Suite

Modular PyTorch framework for benchmarking CNN architectures (ResNet-18, MobileNetV2, GoogLeNet, AlexNet) with grid search over optimizers and hyperparameters.

PyTorchPython
View code
C

Clinical Risk Prediction

End-to-end binary classification pipeline for clinical risk detection, with SMOTE class-imbalance correction and ROC-AUC/confusion-matrix evaluation.

PyTorchScikit-learn
View code
View all projects on GitHub
My technical toolbox

Skills & Tools

Languages

PythonC++JavaSQL

Machine Learning

PyTorchTensorFlowScikit-learnOpenCVMediaPipeTransformers

Generative AI

LLMs (GPT, BERT)RAGLangGraphStable DiffusionGANs

Backend & Data

FastAPIREST APIsPostgreSQLMongoDBRedisElasticsearch

MLOps & Cloud

DockerKubernetesTriton Inference ServerAWSGoogle CloudGitHub ActionsLinux

AI Coding Tools

CursorClaude CodeGitHub CopilotChatGPTGemini
Notes from the build

Articles & Writing

Designing Production-Grade RAG PipelinesRAG
Jun 20, 202611 min read

Designing Production-Grade RAG Pipelines

Lessons from shipping a RAG-powered recipe feature: chunking strategy, hybrid retrieval, grounding and keeping it fast at scale.

Read article
Vector Search in Production with ElasticsearchSearch
Jun 5, 202610 min read

Vector Search in Production with Elasticsearch

How transformer embeddings and Elasticsearch combined to lift search relevance by 35–45% — indexing, ANN tuning and evaluation.

Read article
MLOps at Scale: Serving 20M Requests a DayMLOps
May 22, 202612 min read

MLOps at Scale: Serving 20M Requests a Day

Architecture notes on FastAPI, Docker and Kubernetes for high-throughput model serving with 99.5% uptime.

Read article
Deploying Models with Triton Inference ServerInference
May 8, 202610 min read

Deploying Models with Triton Inference Server

A practical guide to model repositories, dynamic batching and getting sub-3-second response times in production.

Read article
Real-Time Computer Vision on Mobile with MediaPipeComputer Vision
Apr 24, 20269 min read

Real-Time Computer Vision on Mobile with MediaPipe

Integrating MediaPipe with Unity and tuning for mid-range devices — a 40% latency win.

Read article
Building a Reusable Stable Diffusion SDKGenAI
Apr 10, 202610 min read

Building a Reusable Stable Diffusion SDK

How a Python SDK cut feature-hosting time by 80% for a production AI art generator.

Read article
Giving LLM Agents Persistent Memory with pgvectorLLM Agents
Mar 27, 202611 min read

Giving LLM Agents Persistent Memory with pgvector

Session-based semantic memory for multi-turn agentic workflows using PostgreSQL + pgvector and SQLAlchemy.

Read article
97% Test Coverage: CI Pipelines for ML ServicesMLOps
Mar 14, 20269 min read

97% Test Coverage: CI Pipelines for ML Services

Designing GitHub Actions pipelines that cut production errors by 95% for a high-traffic ML platform.

Read article
Code Review with RAG: Building MergeWiseRAG
Feb 28, 202610 min read

Code Review with RAG: Building MergeWise

Combining FAISS-based retrieval with LLM reasoning to review pull requests automatically.

Read article
Have a project or role in mind? Drop me a line.

Let's Work Together

Prefer email?sameetassadullah744@gmail.com
Response timeUsually within 24 hours