Ali Kalakech

University of Windsor — Computer Vision

• –Co-authored Insectopedia: An On-Device Mobile Pipeline for Fine-Grained Agricultural Pest Identification, accepted to the 12th International Workshop on Mobile Applications (IWMA 2026).
• –Built the full inference pipeline to run on-device with zero network dependency, combining YOLO detection, RepViT-SAM segmentation, and group-specific ResNet18 classification.
• –Curated and labelled an 11,768-image dataset spanning 20 species across 9 taxonomic groups.

Contract — Self-employed

• –Owned 20+ projects end-to-end for international clients: requirements and architecture through CI/CD deployment, monitoring, maintenance, and incident response.
• –Built production systems in Go and TypeScript serving thousands of concurrent users.
• –Shipped prototypes and betas early to gather real feedback before committing to larger builds, and handled live incidents under operational pressure.

A production ingestion service that replaced an expensive third-party API dependency for a platform serving 30,000+ daily users across 100+ communities.

• –Identified and solved a USD 5,000+/month API cost problem by reverse-engineering the provider's private GraphQL API — mapping its query and response structures and writing parsers for likes, retweets, and replies.
• –Rewrote the original TypeScript implementation in Go with GoFiber, cutting response times by roughly 40%.
• –Migrated a live production monorepo by refactoring upstream consumers with zero downtime.
• –Designed SOCKS5 proxy rotation, token-pool management, and distributed request handling to keep throughput stable.

• Go
• GoFiber
• TypeScript
• GraphQL
• SOCKS5

An on-device mobile pipeline for identifying agricultural pests in field conditions where the network is unreliable or absent. Accepted to IWMA 2026.

• –Achieved 86.09% species-level accuracy on 1,524 held-out field images.
• –Chained object detection, segmentation, and group-specific classification entirely on-device, with no network round trip.
• –Added confidence-based routing, resolution retries, and two-pass tiling to survive real field-deployment constraints.
• –Added human-in-the-loop correction logging for difficult cases, working toward an approximately 160-species Ontario target.

• Flutter
• YOLO
• RepViT-SAM
• ResNet18
• PyTorch

A self-serve applicant portal that collapsed a complex manual onboarding process into login, submission, automated processing, and administrative review.

• –Built and shipped end-to-end: OAuth authentication, multi-step submission workflows, and an RBAC administrative dashboard.
• –Added background job workers for asynchronous workflow automation.
• –Implemented CI/CD with GitHub Actions, then managed cloud deployment, monitoring, and post-launch incident response.
• –Enabled 4,000+ users and contributed to more than USD 150,000 in client revenue through automated onboarding.

• TypeScript
• Express
• MongoDB
• React
• OAuth
• RBAC

A personal production environment I've run for five years — the place most of my infrastructure and operations experience actually comes from.

• –Proxmox virtualization running 25+ self-hosted services across Linux VMs and containers.
• –OPNsense networking with mesh VPN access, split-horizon DNS, internal SSL, and reverse proxies.
• –GPU passthrough for local AI inference, plus custom agent tooling built on top of it.
• –Backups, recovery, monitoring, upgrades, and the incident recovery that comes with all of it.

• Proxmox
• Docker
• OPNsense
• NGINX
• Linux