Orange Cyberdefense
33.57°N 7.58°W
3+ years delivering secure, scalable systems across cybersecurity, fintech and SaaS. Architect of Dentist365 (1K+ users). Elastic Gold Contributor. Upwork Top-Rated. Now: AI agents and workflow automations that actually ship.
I design for the day you have 10× the traffic, tenants, and teammates. Isolation boundaries, pub/sub topology, and cache layers are decided before the first endpoint is written.
A 40% processing-time cut is not a vibe — it's a before/after benchmark. I profile, remove lazy-loads, rewrite the query, and ship the number with evidence.
LLMs are control flow now. I build agents that observe, plan, call tools, and report back — with guardrails, regression evals, and cost budgets built in from day one.
A PR should read like an essay: tight scope, argued tradeoffs, reversible decisions. A system should read like a diagram — obvious at a glance, defensible under load.
Orange Cyberdefense
Cut API processing time >40% — profiled the hot path, rewrote role/team query logic, shipped with benchmarks.
Designed and enforced multi-tenant data isolation boundaries for enterprise client onboarding.
Automated threat-detection workflows by wiring Laravel schedulers into the Cato security platform.
Architected a pub/sub event bus enabling real-time, decoupled data exchange across internal applications.
Built a skills-aware auto-assignment engine that routes bookings and orders without human dispatch.
LLMs aren't a feature — they're a runtime. I build agents that plan, call tools, and close loops; evals that keep them honest; and automations that turn one-off work into compounding leverage.
Planner/executor agents with typed tool manifests, persistent memory, and retry budgets. Regression-tested and shipped to prod — not just demoed.
Event-driven pipelines that turn raw scrapes, inbound emails, and webhook fires into closed-loop business processes — zero human in the loop.
Hybrid vector + keyword search on Elasticsearch and pgvector. Answers are grounded in source documents, cited, and fast enough to feel instant.
Prompt versioning, regression eval suites, cost dashboards, and safety rails. Prompts live in version control, not in someone's head.
Drop-in widgets and MCP-compatible servers that bring AI capabilities into the surfaces users already work in — no new app required.
Structured, schema-validated output from messy PDFs, scraped HTML, and raw transcripts. If it can be read, it can be parsed.
This isn't a form. It's a live agent grounded in my work history. It reads the questions, reasons, and answers with citations into real projects.
Full practice management — Stripe billing, real-time chat, appointment scheduling, and an embeddable booking widget clinics drop into any site. Grew to 1K+ active users.
C++/Python edge firmware piping sensor data into Kafka → Spark Streaming → BigML. Real-time fall prediction with on-device alerting.
React + Node.js + Elasticsearch. IMDb data ingested and indexed in real time; sub-100ms search across cast, crew, and metadata.
Scrapers pull reviews from major platforms; workflows triage, respond, and escalate — no manual queue for the agency team.
Distributed ingestion pipeline with per-source rate limiting, Elasticsearch indexing, and GitHub Actions orchestrating the daily pricing run.
Users subscribe, upgrade, and cancel from slash commands — no external dashboard, no support tickets. Per-seat billing handled entirely in-channel.