ARYAN SAKHALA
██████╗ ██████╗ ██╗ ██╗ █████╗ ███╗ ██╗██╔═══██╗██╔══██╗╚██╗ ██╔╝██╔══██╗████╗ ██║███████║██████╔╝ ╚████╔╝ ███████║██╔██╗ ██║██╔══██║██╔══██╗ ╚██╔╝ ██╔══██║██║╚██╗██║██║ ██║██║ ██║ ██║ ██║ ██║██║ ╚████║╚═╝ ╚═╝╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝╚═╝ ╚═══╝ ███████╗ █████╗ ██╗ ██╗██╗ ██╗ █████╗ ██╗ █████╗ ██╔════╝██╔══██╗██║ ██╔╝██║ ██║██╔══██╗██║ ██╔══██╗███████╗███████║█████╔╝ ███████║███████║██║ ███████║╚════██║██╔══██║██╔═██╗ ██╔══██║██╔══██║██║ ██╔══██║███████║██║ ██║██║ ██╗██║ ██║██║ ██║███████╗██║ ██║╚══════╝╚═╝ ╚═╝╚═╝ ╚═╝╚═╝ ╚═╝╚═╝ ╚═╝╚══════╝╚═╝ ╚═╝
Security-minded engineer, building agentic AI.
Graduate researcher at UNB CIC. Before that, I shipped RAG and agent systems for Dell and Intel.
# ~/.config/aryan/profile.yamlname: "Aryan Sakhala"role: "Master of Computer Science"program: "Cybersecurity & Agentic AI"affiliation: "UNB CIC (Fully Funded)"location: "Canada"focus:- "Cybersecurity & Agentic AI"- "RAG & LLM Architecture"- "Distillation, mROPE, Parallelism"clients:- name: "Dell Technologies"type: "Enterprise AI"- name: "Intel Corporation"type: "CPU Optimization"- name: "Springer Publications"type: "Research"stats:years_experience: 5projects_delivered: 10+team_members_led: 7publications: 1supercompute_2024:event: "SC24 Demo"project: "Multi-Modal AI Showcase"status: "featured"
ARYAN SAKHALA - System Information===================================OS: MCS @ UNB CICUptime: 5+ yearsShell: /bin/engineerTerminal: Cybersecurity & Agentic AIPROCESSES RUNNING:PID NAME CPU MEM001 cybersecurity 95% HIGH002 agentic-ai 92% HIGH003 llm-architecture 88% HIGH004 research 85% MEDNETWORK:UNB CIC [CONNECTED]Dell Technologies [CONNECTED]Intel Corporation [CONNECTED]
From data science to agentic AI.
From Data Science to SDE II, now a fully funded MCS at UNB CIC — cybersecurity and agentic AI.
Master of Computer Science
University of New Brunswick / CIC — Cybersecurity & Agentic AI
Fully Funded Graduate Research Program | CIC Affiliated
Team Lead, Metrum Insights | RAG Agentic Workflows for Dell & Intel | SC24 Demo
AI HR Bot Development | Dashboard API Pipeline
Team Lead - Proofify (Blockchain) | IUDX Analytics (Gov. India)
Master of Computer Science
UNB | Fully Funded | CIC Affiliated | May 2026 – Present
Every skill, tied to a real project.
Hover a tech tag to see where and how I used it.
Agentic RAG System
Enterprise document retrieval lacked contextual understanding, leading to irrelevant search results across large corpora.
Multi-agent orchestration with distributed task queues, vector similarity search, and LLM-powered reasoning chains.
Production deployment for enterprise clients with async pipeline processing
Contextual retrieval accuracy improved significantly with sub-second query response times across large document sets.
Computer Vision Benchmark Platform
No standardized way to benchmark CV inference across CPU (AMX/non-AMX) and GPU backends with real-time concurrency scaling.
Dockerized multi-backend system — OpenVINO async workers (AMX/CPU), GPU HTTP inference server, Celery coordinator, with real-time metrics via Prometheus.
Tested on Dell PowerEdge R770 (256 cores, dual NVIDIA RTX PRO 6000) processing concurrent video streams at scale
Identified and fixed critical memory leaks (5000-frame replication bug), oneDNN logging disk exhaustion, and Celery visibility timeout infinite-loop — enabling reliable multi-hour benchmark runs.
QAT Hardware Crypto Acceleration
TLS handshake throughput bottlenecked by CPU-bound RSA/ECDSA signing — limiting connections per second for vector database workloads.
QAT Engine/Provider plugged into OpenSSL → HAProxy TLS termination in front of Qdrant. Rustls PoC via patched async crypto provider for benchmark replication.
4x Intel QAT Gen4 devices on dual-socket Xeon 6760P, achieving 58K+ RSA-2048 signs/s (56x over software)
Successfully replicated Intel's benchmark chart (118% for ECDSA P-384). Built and patched C engine with weak symbol stubs for cross-version compatibility.
BBU-RRH Real-Time Logging System
Telecom BBU-RRH systems generated high-volume logs with no real-time streaming or structured analysis pipeline.
Kafka-based log ingestion with topic partitioning, consumer groups for parallel processing, and structured data transformation.
Real-time streaming across distributed telecom infrastructure
Enabled real-time log monitoring and structured analysis for telecom operations.
Player Performance Analytics
Sports analytics teams lacked automated player tracking and performance quantification from video footage.
YOLO-based detection pipeline with tracking algorithms, feeding into ML models for performance classification and statistical analysis.
Multi-video batch processing with real-time dashboard visualization
Automated player tracking and statistical performance analysis from raw video footage.
STT Benchmarking on Cloud
No reproducible benchmark framework for comparing speech-to-text models across different cloud GPU configurations.
Containerized Whisper inference pipeline deployed on AWS with parameterized GPU selection and automated metric collection.
Multi-GPU cloud benchmarking across various instance types
Reproducible STT performance baselines across cloud GPU tiers.
Production Web Applications
Need for modern, performant web interfaces for AI dashboards, e-commerce platforms, and portfolio showcases.
Next.js with SSR/SSG, React component architecture, TypeScript for type safety, and integrated payment systems.
Production-deployed applications serving real users
Responsive, SEO-optimized web applications with premium design and smooth interactions.