Moiz Masood

AI Engineer specializing in Production Computer Vision and Agentic LLM Systems

Based in: Islamabad, PakistanCurrently: AI Engineer, Elunic AGExperience: 1 year

Computer Science graduate with extensive experience in deploying production-grade AI solutions, including visual inspection models and RAG-based engineering assistants. He has a proven track record of delivering high-impact projects, such as an award-winning cardiac MRI analysis pipeline and hallucination-free robotic reporting platforms.

Experience

AI Engineer

Elunic AGwork2026 – Present

Summary

Designed and deployed AI-powered visual inspection solutions across five enterprise manufacturing projects, owning the complete lifecycle from customer requirements and data collection to model deployment and customer validation. Fine-tuned and evaluated object detection and segmentation models (YOLOv8, YOLOv26, RF-DETR, SAM) on 15,000+ industrial images spanning more than 20 defect classes to meet customer accuracy requirements. Performed customer acceptance testing using industry-standard benchmark components, ensuring delivered models satisfied agreed performance criteria before deployment. Developed features for an internal LLM-powered engineering assistant using tool-calling workflows that automated repetitive engineering tasks and reduced routine workflows by 15-20 minutes.

What I did

Deployed visual inspection models in a cloud environment using office instances.

Results

Deployed solutions across five enterprise manufacturing projects.
Managed 15,000+ industrial images and 20+ defect classes.
Reduced routine workflows by 15-20 minutes through LLM automation.
Processed around 500 images a day depending on customer requirements.

YOLOv8YOLOv26RF-DETRSAMLLMTool-calling workflowsNVIDIA RTX 3000 seriesTailscaleKV cachingJSON schemas

Undergraduate Researcher

TUKL Deep Learning Lab, NUSTinternshipJul 2025 – Jun 2026

Summary

Developed deep learning solutions for cardiovascular MRI analysis using PyTorch and MONAI in collaboration with clinical researchers and radiologists. Benchmarked multiple segmentation architectures to identify the best balance between segmentation accuracy, computational efficiency, and deployment feasibility.

What I did

Evaluated segmentation output using Dice score and frequency selection using top-1 and top-3 metrics.

Results

Utilized a private dataset of 26 low frequency scout scans with 11-13 scans per patient.
Finetuned segmentation models using open-source CMRI data from Sunnybrook and ACDC, totaling around 100 patients each.

PyTorchMONAIDeep Learning

Skills

Technical

Python

YOLO

Git

LangChain

RAG

Prompt Engineering

FastAPI

PyTorch

REST APIs

LlamaIndex

OpenCV

Scikit-learn

MLflow

RF-DETR

SAM

Vision Transformers

LangGraph

PostgreSQL

SQLite

Docker

SQL

FAISS

Linux

Hugging Face

Pydantic

C/C++

TensorFlow

MONAI

MySQL

ElasticSearch

Azure

CI/CD

BeautifulSoup

Bash

TypeScript

Neo4j

Next.js

KV caching

JSON schemas

ONNX

Transformers

bitsandbytes

4-bit quantization

GitHub

Quantization

Tailscale

Projects

Clinical Decision Support System for Cardiac MRI (Final Year Project)

Developeracademic

Delivered and presented an end-to-end MRI pipeline (LV segmentation + frequency-offset selection) in a live technical demonstration to academic judges and an industry panel. Awarded Best Industrial Project at SEECS Open House 2026 for developing an end-to-end AI pipeline that automated left ventricular segmentation and cardiac MRI frequency-offset selection. Built and deployed the solution as a Dockerized FastAPI service with MLflow-based experiment tracking, achieving 81.73% IoU for segmentation and reducing MRI frequency-offset selection time from approximately 10 minutes to under one minute.

PyTorchFastAPIDockerMLflowVision TransformerAttention U-Net

MCQ Generation RAG Pipeline

personal

Built a RAG pipeline to generate MCQs from course book material using LangChain for pipeline abstraction and LlamaIndex for data ingestion and chunking.

LangChainLlamaIndexhttps://github.com/moiz56/rag_mcq_generator

LLM as a Judge Semester Project

personal

Built an LLM as a Judge system using open source Llama models. Used 8-bit quantized versions to enable timely inference on local machines with only CPUs.

LlamaQuantizationTransformers

Mixture of Experts (MoE) Lab Work

personal

Implemented a Mixture of Experts (MoE) using a Mistral model for lab work.

bitsandbytesMistral

FacilityOps Report - Robotic Inspection Reporting & Agentic Analytics Platform

Lead Architect/Developerpersonal

Built a PDF reporting engine for robotic inspection that renders JSON run records with RGB and thermal evidence into branded reports and machine-readable manifests, surfacing missing evidence and sensor gaps instead of hiding them. Architected a three-agent LLM layer over a shared 'extended record': code-only derivations (thresholds, group means/max, rankings, run diffs) feed an analytical agent (question → entity router → guarded read-only SQL → prose), a template-driven narrative agent, and a deterministic action-plan agent, all returning a common versioned envelope. Eliminated numeric hallucination by design: the LLM never writes a number. Code fills every value into cited template slots, structured-output schemas constrain model replies, and a verification layer traces every numeral, number word and ordinal back to the records, refusing the output (REFUSED_UNVERIFIABLE) or abstaining when a value cannot be traced. Built evaluation suites with pytest: 30 acceptance tests (TA-01-TA-30) checking rendered PDFs and manifests, per-derivation question suites (D1-D8), and a 20-question canonical set with paraphrases run against the live model, covering correct answers and correct abstentions, with every case saving its answer PDF and full model/SQL trace for debugging.

PythonClaude APIGemini APISQLiteJinja2WeasyPrintPyYAMLpytestLLM AgentsSQLLangGraphhttps://github.com/moiz56/facilityopsJSON schemas

Education

BS Computer Science

National University of Sciences and Technology (NUST)Computer ScienceGPA 3.56Sep 2022 – Jun 2026

Thesis. Clinical Decision Support System for Cardiac MRI

Coursework

Machine LearningDeep LearningLarge Language ModelsData Structures & AlgorithmsProbability & Statistics

Recognition

Awards

Best Industrial Project

SEECS Open House 20262026

Awarded for developing an end-to-end AI pipeline that automated left ventricular segmentation and cardiac MRI frequency-offset selection.

Merit Scholarship

NUST

Recipient for academic excellence.

Best Across Three Cambridge International A Levels

Cambridge International

2nd in Punjab (Top 0.01% globally).

Social impact & activities

Live Technical Demonstration

SEECS Open House 2026other2026

Presented an end-to-end MRI pipeline to academic judges and an industry panel.

Hire MoizReach out about a role, a contract or a conversation.For recruiters

Moiz's twin is AI, it can make mistakes.

More people on Kanvis

Browse more people Create your own Kanvis