Faisal Basha

• Deployed and optimised large language models in production — Qwen 30B / 235B, GPT-OSS, Llama 3 / 4, Mistral — using vLLM for
high-throughput inference, GPU partitioning (MIG, time-slicing), tensor / pipeline parallelism, and quantisation (AWQ, GPTQ, FP8) to
balance latency, throughput, and cost.
• Built GPU-accelerated ML and computer-vision pipelines (CUDA, PyTorch, MLflow) — YOLO for real-time detection, ResNet / EfficientNet /
Vision Transformers for image classification and feature extraction, and TensorRT / quantisation for low-latency edge deployment on Jetson
and Raspberry Pi.
• Designed production RAG pipelines (LangChain / LangGraph, Pinecone, FAISS) with hybrid retrieval and semantic chunking; deployed via
Docker / Kubernetes / FastAPI.
• Conducted research and prototyping on NVIDIA H100 and A100 GPUs (via Colab Pro and self-managed instances) — including Graph Neural
Networks (GNNs) for relationship-aware reasoning, and explainable-AI frameworks (SHAP, LIME, Integrated Gradients, Captum) for model
interpretability and bias evaluation.
• Operated end-to-end MLOps platform on Azure AKS with GPU node pools — MLflow for experiment tracking and model registry, Apache
Airflow for training-pipeline orchestration and scheduled retraining, and GPU-aware autoscaling for cost-efficient training and serving
workloads.
• Provisioned and operated production AWS EKS clusters via Terraform — IRSA for pod-level IAM, Karpenter and Cluster Autoscaler for
cost-aware node scaling, AWS Load Balancer Controller for ingress, External-DNS, and Velero for cluster backup and DR.
• Authored Ansible playbooks and roles for fleet-wide configuration management — CIS-benchmark OS hardening, patch automation, kernel
tuning, and zero-touch application deployment across hybrid Linux estates (Ubuntu, RHEL, CentOS) on AWS EC2 and on-prem.
• Hardened container build pipelines with multi-stage Dockerfiles, distroless / minimal base images, Trivy and Anchore vulnerability scanning,
and ECR lifecycle policies — reducing image footprint and CVE exposure across services.
• Standardised application packaging via Helm chart libraries with environment overlays; integrated with ArgoCD GitOps for
multi-environment promotion, drift detection, and automated rollback.
• Designed ELK stack architecture with Filebeat / Fluent Bit log shippers, Logstash enrichment pipelines, and ILM-driven hot / warm / cold
tiering for cost-efficient log retention; built Kibana dashboards for SecOps, SRE, and audit workflows.
• Implemented secure CI/CD on Jenkins and AWS CodePipeline / CodeBuild / CodeDeploy with embedded SAST, dependency and container
scanning; enforced K8s RBAC, image hardening, and network policies.
• Enabled Zero-Trust architecture via mTLS (Istio), identity-based access controls, and secure API gateways across microservices; centralised
secrets in HashiCorp Vault and AWS Secrets Manager.
• Implemented Prometheus, Grafana, and AlertManager observability with SLO-based alerting; integrated DAST workflows, runtime container
security, and structured RCA reporting for HA workloads.

https://este-project.com/

Working as a IT Manager, which involved me in managing and maintaining function Apps, Logic Apps, creating & querying tables & Stored Procedures in all environments such as development, staging & production, and making decisions for optimized running of resources in azure portal to support the day today working process of Call Center Process (BPO) for our client company named jobFIXers.

Responsibilities:
•Creating Azure functions which accepts data and filters data from SQL SERVER 2017, which initiates stored procedures.
•Creating Logic Apps that runs recurrence scheduler to change status of persons in database stored in SQL Server 2017 based on certain conditions.
•Creating Azure Table Storage to keep track of events triggered.
•Build Secure & Management of Restful API to read, write, update & delete data from / to function App, logic App etc. etc.
•CI/CD pipeline features to manage & deploy different version of releases for development, staging & production.
•Azure Machine Learning Studio used to extract keywords from database tables and optimize search results to find persons with similar work niche.
•TDD & BDD principles where applied to create Unit Tests & End to End Tests.

• Provisioned AKS clusters via ARM and Terraform with VNETs, NAT gateways, and VNET endpoints for Azure Storage / SQL / ACR.
• Deployed Jenkins on AKS via Helm with internal Azure load-balancer for private CI/CD access.

• Built multi-language CI/CD (Python, C/C++, NodeJS, VueJS) on Jenkins with SonarQube, JFrog Artifactory, and Docker integration.
• Provisioned on-premises Kubernetes clusters (kubeadm on Ubuntu / CentOS) and deployed containerized applications via Ansible-driven
pipelines.
• Configured LDAP across Jenkins, Jira, Confluence, and BitBucket; deployed Snipe-IT, SonarQube, and NGINX SSL via Ansible playbooks.
• Built ML-based network-traffic congestion monitoring service (Python) for proactive traffic prediction on production infrastructure.

• Built GPU-accelerated computer-vision pipelines (CUDA, PyTorch) with MLflow
experiment tracking; deployed YOLO models optimized via TensorRT and quantization for
low-latency inference on Jetson / Raspberry Pi edge devices.
• Designed production-grade RAG pipelines integrating LLMs (OpenAI, Hugging Face)
with vector DBs (Pinecone, FAISS) for enterprise knowledge retrieval and context-aware
Q&A.
• Orchestrated multi-agent LLM workflows with LangChain / LangGraph; engineered
prompt strategies that reduced hallucinations and cut token cost on production workloads.
• Owned end-to-end model lifecycle — data validation, feature engineering, distributed
training, evaluation harnesses, controlled rollouts, and post-deployment drift / quality
monitoring.
• Engineered semantic search pipelines with hybrid BM25 + vector retrieval and
cross-encoder re-ranking; deployed scalable GenAI services via Docker / Kubernetes /
FastAPI.
• Designed cost-aware GPU scheduling and inference auto-scaling on EKS / AKS,
balancing throughput SLOs with FinOps spend targets across training and serving
workloads.
• Implemented LLM observability (latency, token usage, response quality) and AI
guardrails — prompt-injection mitigation, output validation, Responsible AI compliance.
• Integrated GenAI capabilities into real-time, event-driven architectures using Kafka
and microservices.
• Built CI/CD for scientific-ML workflows (Julia PDE models); containerized via Docker
and ran distributed training on Kubernetes with full reproducibility via MLflow.
• Provisioned HPC / AI environments using Terraform and Ansible for large-scale PDE
simulations and ML workloads; profiled and optimized CPU / GPU utilization.
• Automated data pipelines for simulation inputs/outputs to handle large-scale numerical
datasets and model artifacts.
• Designed end-to-end ML pipelines on AWS SageMaker covering preprocessing, training,
hyperparameter tuning, and scalable inference endpoints.
• Built distributed GPU-based training on EC2 with CUDA; deployed real-time and batch
inference via Lambda + API Gateway.
• Automated data ingestion and feature engineering using S3 and AWS Glue for
large-scale ML datasets.
• Provisioned AKS clusters via ARM templates and Terraform; managed application
lifecycle through Helm packages.

• Provisioned Cisco UCS infrastructure (VLANs, VNICs, IP/MAC pools) and on-prem Kubernetes (kubeadm) across DEV / UAT / PROD
environments.
• Managed OpenShift cluster scaling on AWS EC2; deployed 100+ microservices in Python, Java, Angular, and .NET Core.
• Set up private Docker registries, MetalLB load balancing, and NGINX-Ingress for routing across environments.

Creating Reports and R language / R Studio Charts for the Meteorological System at Sharjah International Airport

Created the following reports:
•Values at Observation Time.
•Daily Means and Extremes.
•30 Minutes Daily Statistics.
•AFTN Report Selected.
•Instant value at 30 minutes’ interval.
•AFTN Report for the past 24 hours.
•Instant value for the last 30 minutes.
•Observation value for 24 hours.
•Wind gust 24 hours
•Daily Means and Extremes 1
•Daily Means and Extremes 2

Each of the reports required simple calculations such as mean, median, minimum, maximum and instant values for some time. We used SAP Crystal Reports to make the above reports, which imported data from Oracle Database 11g. The next step was export of data to file formats such as CSV, XLSX, XLS, and PDF. We used R language and R Studio to create Annual wind rose, which takes csv data and converts it to a chart.

Design and Development of Weather Reporting Software
Software Development and Design of Meteorological Reporting System, which required us to work in team of four following the agile pattern. Our roles and responsibilities involved data handling coming from the sensors such as temperature, pressure etc. Calculating values such as QFE, QFF, WB, Wind Speed Wind Direction etc. Saving the data in an RDBMS such as Oracle 11g and finally exporting as a summary in many file formats. We can now query reports within a given time period specified in a calendar query field. We used Crystal Reports via Visual Studio C# for making front-end design. InstallShield software to create the setup and installation files for packaging and installing into other machines.


Responsibilities:
•Troubleshooting of oracle database and software development of flight information display
•Troubleshoot and solve local-area network problems; provide technical support to end-users regarding computer hardware and software; install, test, certify and troubleshoot networking cabling systems
•Performing maintenance and troubleshooting of all IT systems and equipment at Sharjah International Airport, which includes faultfinding and rectification. Making regular property inspections and check lists.
•Follow up and repair outstanding faults and performing day shift or different shift duties and non-scheduled tasks and preventive maintenance of IT systems and equipment.
•Daily basis reporting to supervisors. Receiving the materials and checking the quality. Responsible for corrective and preventive maintenance of IT system.

Developed Windows Phone 8.1 Application namely NMInformatics, AP3 now called PRSY (personalized application), MEMC, SCHS. MEMC was a mobile application for IOS, Android and Windows phone 8.1, which provide users with service to schedule appointments with different doctors. SCHS was a mobile application for Sharjah humanitarian centre, Teaching and training for people on all mobile platforms.

Responsibilities:
•Back end coding for Connectivity of Android Application with Server and Database, SQLite, Webview, push notification, broadcast receivers, intents, async tasks, and singleton objects.
•Built user interface components and added functionality to those components.
•Automation of IOS application from archiving to the deployment in the iTunes application store using fastlane tools.
Technologies: Objective-C, iPhone SDK, XCode, Instruments, gdb Debugger

Taught IOS / Objective-C for master’s Students at the University of Vels Located

mindstorm-presidents.appstor.io/
Worked in an application which is a corporate customer transaction facility created for the RBS customers to do payments. The Project covered addition of new enhancements to the application like financial transfers, Inter Account Transfers (IAT) and development of the same.
Responsibilities:
•Front End Design: Designing each page in Cocos2D
•Built user interface components and added functionality to those components.