Nihad Askri - AI Engineer

Overview

Medical imaging ML engineer (senior-level) focusing on 2D-3D X-ray registration, learned pose estimation, segmentation and related CV tooling. The strongest proven skill is building and evaluating model-based and classical registration workflows, grounded in artifacts such as src/registration/pose_cnn.py and src/registration/compare_classical_vs_learned.py. Public code shows limited MLOps productionization and minimal formal experiment tracking, and many infrastructure/serving files were marked as boilerplate rather than original work.

Phone

Technical skills

Languages

4

Python

SQL

JavaScript

MATLAB

AI/ML

28

LLM

RAG

Machine Learning

NumPy

OpenCV

Scikit-learn

Datasets

Transformers

LangChain

LangGraph

LoRA

YOLO

ChatGPT

ONNX Runtime

PyTorch

CatBoost

PEFT

Qwen

Whisper

OpenAI

Anthropic

Hugging Face

MLFlow

vLLM

CLIP

DINOv2

Time Series

Model Context Protocol

DevOps

8

GCP

Grafana

GitHub

Docker

GitLab

GitLab CI

Linux

Prometheus

Other

18

React.js

Qdrant

Matplotlib

Uvicorn

AI Agents

CI/CD

Jenkins

LiveKit

Deep Learning

Classic ML

Quantization

Computer Vision

ETL/ELT

LLM Guardrails

Multimodal AI

NLP

Speech Recognition

Image Segmentation

Projects

Multimodal LLMs

Fine-tuned multimodal large language models for understanding across text, audio, and video. Used Hugging Face transformer tooling and parameter-efficient fine-tuning to adapt model behavior. Supported experiments focused on multimodal inputs and task-specific adaptation.

Multimodal Argumentation & Fallacy Detection

Implemented multimodal models for argumentative fallacy detection by combining language and additional input representations. Trained and evaluated the approach using transformer-based encoders and cross-attention mechanisms. Focused on improving detection quality for argumentative structures.

Multimodal Conversational AI

Developed a real-time conversational pipeline for low-latency speech interaction. Combined speech-to-text via Whisper, end-of-utterance detection, and a live audio streaming component. Focused on integrating the components into a cohesive system for conversational behavior.

LLM Gateway

Created a FastAPI-based gateway to route requests to local or cloud LLM endpoints. Added ML-driven routing, guardrails, and rate limiting while tracking cost and latency. Implemented observability with metrics to support monitoring and debugging of the system.

Guarded Agentic RAG

Built a PDF-based agentic RAG system using an agent workflow with multi-tool reasoning. Implemented document ingestion with sentence-aware chunking, embedding and retrieval, and evaluation of generated responses. Added protections against prompt-injection attacks to improve robustness of the pipeline.

Timeline

AI Research Engineer · Middle

LISSI Laboratory of UPEC · Full-Time

Mar 2026 to May 2026 2 Months Mila In office

Developed and evaluated multimodal deep learning and representation learning models for research use. Built datasets via ETL, quality checks, and feature/statistical analysis, then ran experiments with modern transformer-based architectures. Focused on computer vision and multimodal model training and benchmarking.

Transformers

DINOv2

CLIP

ETL/ELT

Part-time Instructor (Deep Learning and AI) · Head+

University Center of Abdelhafid Boussouf · Part-Time

Jul 2025 to Sep 2025 2 Months In office

Delivered practical deep learning sessions to master’s-level students and taught an introductory AI course for first-year students. Prepared and facilitated coursework to support students in understanding core concepts and applying them in hands-on settings.

Final-year Intern (AI/ML) · Junior

LISSI Laboratory of UPEC · Internship

Oct 2023 to Mar 2025 1 Year 5 Months Vitry-sur-Seine In office

Developed AI solutions spanning data preparation, model development, evaluation, and deployment, combining vision, language, and speech through LLM and RAG pipelines. Worked on continual learning and time-series machine learning for video-based activity recognition and user behavior prediction. Contributed to end-to-end experimentation workflows from training to deployment.

LLM

RAG

Time Series

Full-stack Developer · Middle

Freelance · Freelance

Apr 2023 to Sep 2023 5 Months Vitry-sur-Seine In office

Built an end-to-end web application from the front end to the back end and connected it to a database. Developed user-facing interfaces and server-side APIs using a JavaScript stack. Delivered a working product suitable for real use rather than prototypes only.

React.js

Node JS

Final-year Intern (Machine Learning) · Junior

LCSI Laboratory of ESI · Internship

Jan 2023 to May 2023 4 Months Paris In office

Worked on a machine learning-based approach to predict code refactoring in Java. Contributed to developing and evaluating the model components for the prediction task. Focused on applying ML/DL methods within a software-engineering context.

Java

Machine Learning

AI Team Lead · Lead

Sigservice · Full-Time

In office

Led a production-ready video recognition pipeline covering end-to-end processing, feature embedding generation, and embedding retrieval. Built recommendation and sales forecasting models using classical ML and sequential recommendation approaches. Worked on system integration aspects to support deployment of AI components in production.

YOLO

Qdrant

CatBoost

GCP

Junior AI/ML Engineer Confidence: Medium ML Engineer

Medical imaging ML engineer (senior-level) focusing on 2D-3D X-ray registration, learned pose estimation, segmentation and related CV tooling. The strongest proven skill is building and evaluating model-based and classical registration workflows, grounded in artifacts such as src/registration/pose_cnn.py and src/registration/compare_classical_vs_learned.py. Public code shows limited MLOps productionization and minimal formal experiment tracking, and many infrastructure/serving files were marked as boilerplate rather than original work.

Model Architecture & Training

5/10

How well models are designed and trained

Custom model architectures and training loops for pose estimation and segmentation are implemented, showing practical ML engineering and model-design experience.

Evidence

askrinihad/xray-2d3d-registration/src/registration/pose_cnn.py: class PoseCNN askrinihad/xray-2d3d-registration/src/registration/train_pose_cnn.py: training loop and optimizer setup askrinihad/xray-2d3d-registration/src/segmentation/unet.py: SmallUNet implementation

Data Pipeline & Feature Engineering

4/10

How data is prepared for models

Data loading, synthetic sequence generation and DRR rendering for medical imaging are present, indicating applied data-pipeline work and domain-specific augmentation/data synthesis.

Evidence

askrinihad/xray-2d3d-registration/src/segmentation/dataset.py: ArcadeSegmentationDataset dataset class askrinihad/xray-2d3d-registration/src/tracking/generate_synthetic_sequence.py: synthetic sequence and deformation_field askrinihad/xray-2d3d-registration/src/registration/common.py: load_subject, build_camera, sample_random_poses

Experimentation & Evaluation

4/10

How results are measured and tested

There are evaluation and comparison scripts that validate learned vs classical methods and compute metrics, but formal experiment tracking or reproducible run history is limited.

Evidence

askrinihad/xray-2d3d-registration/src/registration/compare_classical_vs_learned.py: evaluation and classical vs learned comparison askrinihad/xray-2d3d-registration/src/segmentation/evaluate.py: segmentation evaluation harness

MLOps & Deployment

How models are shipped to production

Not evidenced in public code

Computational Efficiency

2/10

How efficiently computing resources are used

Some attention to practical runtime and numeric work appears in training scripts and synthetic data choices but there is little original GPU/quantization/benchmark engineering visible in the human-authored code.

Evidence

askrinihad/xray-2d3d-registration/src/registration/train_pose_cnn.py: explicit training loop and use of torch for data/model execution askrinihad/xray-2d3d-registration/src/registration/common.py: utilities for pose normalization and sampling which affect batching/efficiency

Research Depth & Innovation

4/10

Depth of research and new ideas

Implements and compares a classical gradient-based registration baseline with a learned CNN and uses a domain-specific differentiable renderer, showing research-minded implementation and problem framing.

Evidence

askrinihad/xray-2d3d-registration/src/registration/compare_classical_vs_learned.py: explicit classical gradient descent registration baseline askrinihad/xray-2d3d-registration/src/registration/generate_drr.py: uses nanodrr to render DRRs for experiments askrinihad/xray-2d3d-registration/src/registration/pose_cnn.py: custom pose-prediction CNN

Expertise

Computer Vision & Image Analysis· Junior

Medical AI & Healthcare· Junior

Industries

Health Care· Junior

Technologies

SQL

MATLAB

LangGraph

LangChain

Qwen

CatBoost

GCP

ChatGPT

Qdrant· since 2022

LoRA

Model Context Protocol

vLLM

MLFlow

YOLO

Prometheus

Quantization

Multimodal AI

Computer Vision· since 2026

AI Agents

NLP· since 2026

Speech Recognition

GitLab CI

PEFT

CI/CD

Jenkins

Transformers· since 2026

PyTorch· since 2026

Docker

Grafana

LLM· since 2023

RAG· since 2023

Whisper

CLIP· since 2026

DINOv2· since 2026

Time Series· since 2023

Image Segmentation

Hugging Face

GitHub· since 2026

GitLab

LiveKit

LLM Guardrails

Linux

Machine Learning· since 2023

Recommendations

Develop end-to-end medical imaging model pipelines that include robust dataset versioning and reproducible experiment tracking.
Build optimized inference artifacts and benchmarks for embedded/real-time targets, including validated quantization and latency measurements under real workloads.
Work on advanced CV components such as multi-view registration, uncertainty estimation, and sensor-specific calibration tooling for clinical integration.
Package tested model deployment flows (CI/CD, containerized inference, monitoring) for constrained devices or hospital edge servers.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

xray-2d3d-registration llm-gateway voice-agent-platform

Intern Data Scientist Confidence: Low Generalist

A junior-level generalist focused on ML prototypes, model deployment scaffolding, and LLM agent tooling. The strongest proven skill is production-aware model deployment and validation, demonstrated by the ONNX export/profile/validate scripts (src/deployment/export_onnx.py, profile_models.py, validate_onnx.py). There is little or no evidence of extensive human-authored system ownership, comprehensive unit/integration tests, or mature cross-project architecture in the provided human-authored artifacts.

Statistical Rigor

Correct use of statistics

Not evidenced in public code

Data Wrangling & Cleaning

Preparing and cleaning data

Not evidenced in public code

Exploratory Analysis & Visualization

Exploring and visualizing data

Not evidenced in public code

Predictive Modeling

Building models that predict

Not evidenced in public code

Business Insight & Impact

Turning analysis into business value

Not evidenced in public code

Reproducibility & Notebook Hygiene

Clean, repeatable analysis

Not evidenced in public code

Industries

Health Care· Junior

Artificial Intelligence· Intern

Technologies

Classic ML

Deep Learning

Python· since 2025 · Junior

OpenCV

Scikit-learn

Matplotlib

Datasets

NumPy

Uvicorn

OpenAI

Anthropic

ONNX Runtime

Recommendations

Consolidate and author core modules with clear ownership and add unit tests for parsing/logic (for example, tests for parse_response and tool-parsing logic).
Add CI and pinned environment specifications (requirements.lock or poetry.lock) and a small reproducible demo with deterministic seeds to improve reproducibility.
Increase human-authored coverage on security and safety checks with unit/integration tests for pre_check/post_check logic and add audit logs for security events.
Iterate on end-to-end RAG evaluation and create reproducible evaluation notebooks or scripts that can be run in CI without separate manual venvs.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

xray-2d3d-registration llm-gateway voice-agent-platform

Intern Backend Developer Confidence: Low Generalist

Backend-focused generalist at an Intern level with early-stage work around LLM and voice-agent concepts. There is no verifiable human-authored artifact in the analyzable files to support a single strongest proven skill; most visible material is dependency lists and files marked as boilerplate or AI-authored. Production-grade backend concerns such as transactional design, migration history, resilience patterns, and audited security controls are not evidenced in the human-authored files provided.

API Design

How well APIs are designed

Not evidenced in public code

Data Layer & Database

Working with databases

Not evidenced in public code

Scalability & Performance

Handling load and speed

Not evidenced in public code

System Architecture

Overall system structure

Not evidenced in public code

Security & Auth

Protecting data and access

Not evidenced in public code

Reliability & Observability

Stability and monitoring

Not evidenced in public code

Recommendations

Assign to prototype features and iterate on well-scoped LLM routing or tool-invocation logic under senior mentorship.
Have the developer implement and unit-test pure parsing and transformation functions (logic that can be fully validated without live API keys) to build reliable, testable components.
Use for documentation, dependency hygiene, and turning high-level designs into small, reviewed PRs while pairing with an experienced engineer for security and resilience decisions.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

llm-gateway voice-agent-platform askrinihad