Abbas Mustafa — AI Systems Engineer & Cybersecurity Specialist
Location: Rennes, Brittany, France |
Email: abbasmsh1@gmail.com |
LinkedIn: linkedin.com/in/abbas-mustafa |
GitHub: github.com/abbasmsh1
AI Systems Engineer with 5+ years of experience designing and deploying secure, autonomous multi-agent AI systems. I bridge the gap between experimental Machine Learning and production-grade MLOps. Currently pursuing an MSc in Cybersecurity at IMT Atlantique (Rennes, France) while working as a freelance AI engineer for international clients in healthcare, legal, and finance sectors.
Skills & Expertise
- AI & Machine Learning: PyTorch, TensorFlow, MONAI, Hugging Face, LangChain, LangGraph, LangSmith, CrewAI, OpenCV, U-Net, YOLO, RAG, Embeddings, Transformers
- Security & Defense: SIEM (Splunk, ELK), IDS/IPS, Network Forensics, Malware Analysis, Sandboxing, Adversarial ML, SOC Automation, Threat Hunting
- Infrastructure & DevOps: FastAPI, Flask, Django, Docker, Kubernetes, AWS, Azure, GCP, PostgreSQL, MongoDB, Redis, Pinecone, ChromaDB, Weaviate
- MLOps & Optimization: MLflow, Weights & Biases, DVC, ONNX Runtime, TensorRT, Mixed Precision Training, CI/CD
- Core Programming: Python (Expert, 5+ years), C++, JavaScript/TypeScript, Go, SQL, Bash
Work Experience
Contract AI Engineer — Freelance (2020 – Present, Remote)
- Architected and deployed 15+ autonomous agent systems for international clients across healthcare, legal, and finance sectors.
- Built production-grade FastAPI microservices serving 100K+ daily ML inference requests with <100ms latency.
- Developed 'CoyChatbot', a dual-LLM conversational agent with RAG, achieving 95% user satisfaction.
- Implemented MLOps pipelines with automated model versioning, A/B testing, and monitoring using MLflow and Weights & Biases.
- Designed secure AI systems with adversarial defense mechanisms and prompt injection protection.
AI / ML Engineer — Orydex (May 2025 – Jul 2025, Remote)
- Fine-tuned U-Net and attention-based models for multi-organ CT scan segmentation, achieving 92% Dice score.
- Optimized PyTorch/MONAI training pipelines with mixed precision and distributed training, reducing latency by 25%.
- Developed ONNX inference microservice for cross-platform deployment (CPU/GPU/Edge devices).
AI Researcher — AIM Lab (Jan 2021 – Dec 2023, Pakistan)
- Led research on Roman Urdu to Urdu Neural Machine Translation using Transformer architectures.
- Developed custom dataset of 50K+ parallel sentences with preprocessing pipeline.
- Published findings 'CMRUTU: Roman Urdu to Urdu Neural Machine Translation' in IEEE FIT Conference 2022.
Projects
Multi-Agent Security Operations Center (Current)
Autonomous 3-tier analyst system (Tier 1, Tier 2, Tier 3) simulating real SOC operations with intelligent escalation logic. Agents autonomously analyze 10,000+ network alerts, correlate threats, and execute IP blocking mitigation. Achieves 95%+ accuracy in threat classification, reducing analyst workload by 70%. Built with LangChain/LangGraph orchestration and MCP server integration.
Technologies: LangChain, LangGraph, Python, Multi-Agent Systems, Cybersecurity
Medical Imaging Microservices Platform (2025)
Scalable microservices platform for medical image analysis with U-Net segmentation models. ONNX inference engine with <100ms response time. Processes 1000+ CT scans daily with HIPAA-compliant data handling.
Technologies: PyTorch, ONNX, Docker, FastAPI, Healthcare AI
AI-Powered Intrusion Detection System (2024)
Neural network-based IDS analyzing 10M+ network flows with 98%+ accuracy. GPU-optimized training with Mixed Precision (AMP) achieving 40% speedup. Research on adversarial robustness of ML-based intrusion detection.
Technologies: Deep Learning, PyTorch, Network Security, GPU Optimization
CoyChatbot — Dual-LLM Conversational Agent (Freelance 2023)
Adaptive conversational agent using dual-LLM architecture. Serving 100K+ daily requests with <100ms latency and 95% user satisfaction. Context-aware with Redis session persistence.
Link: coy-chatbot.vercel.app
Torrential — BitTorrent Client (2026)
BitTorrent downloader built on libtorrent with a CLI (live progress, ETA, rate limiting) and a web client: FastAPI JSON API plus React/Vite SPA managing many torrents with persisted state.
Link: github.com/abbasmsh1/Torrential
Thulla — Multiplayer Card Game (2026)
Real-time multiplayer trick-taking card game for 2–14 players. React + TypeScript frontend, Python/FastAPI backend, gameplay synchronized over WebSockets.
Link: thulla-one.vercel.app
AdsAssistant — Google Ads Strategist Agent (2026)
Senior-level digital marketing strategist agent powered by LangChain and Mistral AI. Analyzes campaigns, search terms, and keywords; generates RSA-compliant ad copy and weekly performance summaries.
Link: ads-assistant-drab.vercel.app
SaffarLog — Travel Management Platform (2026)
Travel management platform with authentication and roles, trip planning, destination discovery, an AI travel assistant chat, and a full admin CMS. Built with Next.js, TypeScript, Prisma, and PostgreSQL.
Link: saffarlog.vercel.app
Education
- MSc in Cyber Security — IMT Atlantique, Rennes, France (2025 – 2027)
Specializations: Network Defense, Secure AI Architecture, SOC Automation
- BS Artificial Intelligence — FAST NUCES, Islamabad, Pakistan (2020 – 2024)
Publications
M. Wisal, A. Mustafa et al., (2022). "CMRUTU: Roman Urdu to Urdu Neural Machine Translation." IEEE FIT Conference 2022.
DOI Link
Awards
- 1st Prize @ SOFTEC 2023 — Stealth Scraper: High-velocity data extraction engine evading modern bot detection using fingerprint randomization.
- 1st Prize @ LAPSX 2022 — Auto Color-Grade: Automated cinematic color grading using computer vision and reference histogram matching.
Blog Posts
- Building a Multi-Agent SOC: From Concept to Production (Nov 15, 2024, 12 min read) — Deep dive into architecting a 3-tier autonomous Security Operations Center using LangGraph.
- Securing LLM Agents in Production (Oct 12, 2024, 8 min read) — Defense layers against prompt injection and jailbreaking in autonomous agents.
- Optimizing Medical Imaging Models for Edge Deployment (Sep 28, 2024, 10 min read) — PyTorch to ONNX: model quantization and <100ms inference for healthcare edge devices.
- GPU Optimization for Deep Learning: Mixed Precision Training (Sep 10, 2024, 7 min read) — 40% training speedup using PyTorch AMP on IDS models.
- Dockerizing GPU Workloads for ML (Aug 22, 2024, 9 min read) — NVIDIA Container Toolkit, CUDA optimization, and Kubernetes deployment for ML pipelines.
- From Student to SOFTEC Winner: Engineering the Stealth Scraper (Aug 15, 2024, 6 min read) — Anti-bot web scraper using fingerprint randomization and behavioral mimicry.
Frequently Asked Questions
- What are Abbas Mustafa's main technical skills?
- Abbas specializes in multi-agent AI systems (LangGraph, CrewAI), adversarial machine learning, cybersecurity, MLOps, and production AI deployments using Python, PyTorch, and FastAPI.
- Is Abbas Mustafa available for freelance work?
- Yes. Contact abbasmsh1@gmail.com or via LinkedIn.
For a complete AI-readable summary, visit https://abbas-mustafa.dev/llms.txt