A decade as the escalation path.I build the systems that end it.
Paris · On-site · Full time
Dust says support is one of the most important product surfaces it owns — I've spent a decade agreeing with that, from the inside, under founder titles. I've been the technical escalation path across nine client geographies and five time zones, and I keep turning the recurring case into a system instead of a repeat ticket: an agentic CMS at , an n8n pipeline at , evaluation harnesses that catch a wrong AI answer before a customer does. I'm relocating to Paris on purpose.
Certified & shipped with
Aptitude. Attitude. Agency.
Dust's own hiring dimensions, mapped honestly — practitioner depth, receipts over adjectives, and a decade of finding the problem before anyone scoped it.
Aptitude
Builder-level AI fluency plus technical depth that doesn't stop at a stack trace — Anthropic's own Claude Code and Agent Skills certifications, an n8n pipeline shipped in production at , and a decade of reading logs, code and product behavior across nine client geographies.
Attitude
Resilience and empathy proven under real stakes, not simulated ones — weeks inside 's clinics where the frustration belonged to a patient, not a ticket — plus a founder's reflex to own the outcome rather than route it upstream.
Agency
Patterns become systems, not repeat tickets. 's CMS turns a non-technical client's plain-English requests into agent-authored, CI/CD-gated production changes; a decade of escalations arrive with evidence and a recommendation attached; six solo AI-native products shipped in the last year because nobody else was going to build them first.
Ask Fauzul's AI
What you're hiring for, running.
Dust structures this role around three dimensions — Aptitude, Attitude, Agency — and says explicitly what it is not: not a ticket-jockey seat, not people-management-from-a-distance, not pure ops, not AI-curious-without-building. Here is the point-by-point mapping of a decade as the escalation path, and the systems he built so escalation stops being necessary, with the real ramps marked honestly.
What's a support system he's actually built, not just used?
Global Jute's CMS: a non-technical client edits their live site in plain English; an agent authors the change; CI/CD gates it to production. The client publishes without waiting on developers.
Build the systems that make support scale
Deflection engineered, not hoped for — plain-English self-serve, agentic pipelines, and eval harnesses that catch bad output before a customer does.
You have experience operating at the strategy layer of support, not only at the ticket layer. You have owned or helped shape support systems, support tooling, escalation workflows, deflection strategy, knowledge management, or AI support operations.
The strategy-layer work is real — he shaped escalation workflows, deflection systems ('s agentic CMS), and knowledge/eval harnesses (, ) — but it happened inside founder and strategic-partner roles (, ), never as a titled support-org seat at a SaaS vendor. He's shaped the strategy layer; he hasn't held the job title.
You turn patterns into systems. When you see the same issue twice, you start asking whether it should exist at all.
's CMS is the receipt: instead of routing every content change through a developer, the client describes the edit in plain English, an agent drafts the change, and CI/CD gates it to production — the client publishes without waiting on developers. It's real deflection engineering, built and shipped solo. What's honestly missing is doing this inside a formal support stack with queue volume, CSAT, or TTR tracked as KPIs — his instinct for turning repeat pain into a system is proven; the metrics discipline around it has come from instrumented pilots and evaluation harnesses, not a ticketing platform.
You build without waiting for permission. If you see a gap in the support system, you are excited to prototype, test, and ship a better version.
Six AI-native products shipped solo in the last year — (built during the World's Largest Hackathon and submitted to Y Combinator), (Milan AI Week 2026, results pending), , the CMS, , and . Nobody scoped or greenlit any of them; he found the gap and built.
How long has he been the escalation path?
A decade, under founder titles — nine client geographies, five time zones, and EKAGRA's clinic floors under real patient-facing pressure.
Run the complex escalations
A decade as the escalation path across nine client geographies and five time zones — logs to code to product to business impact, translated in both directions.
You have strong technical depth. You are comfortable with APIs, web applications, logs, scripts, and internal tools. You can navigate a codebase, reason through product behavior, and troubleshoot systems without being stopped by a stack trace.
Ten years of shipping production TypeScript/Node.js and React, REST/GraphQL/webhook integrations across Stripe, Vipps, and third-party SaaS, and internal tools built and open-sourced (). He reads other people's stack traces for a living — a decade of client escalations across nine geographies means logs, codebases, and product behavior he didn't write are the daily job, not the exception.
You are able to communicate bidirectionally: clearly enough for customers, technically enough for engineers, and strategically enough for leadership.
A decade of altitude-switching: architecture conversations with VPs and C-suite in the morning, pairing with staff engineers in the afternoon, and translating both into terms a non-technical clinic owner or export-house principal could act on. 's owners now run their live site by describing changes in plain English — the same translation instinct this role asks for, aimed at customers instead of clients.
You combine resilience with empathy. This is a front-line role, and customer frustration is often direct and unfiltered. You do not take it personally, but you do take it seriously.
Weeks spent physically inside 's diabetes and wound-care clinics, watching real patients' frustration under real pressure — not a support queue's abstraction of frustration, the thing itself. As a founder, there was never someone else to escalate a client's anger to; resilience and empathy were the job description, not a training module.
What does 'going deep before escalating' look like in practice?
A decade of writing the reproduction-plus-evidence-plus-recommendation escalations his own engineers received — never a bare 'it's broken.'
Turn recurring pain into product signal
Field observation becomes roadmap — the same discipline that turned clinic-floor friction into EKAGRA's product direction and client auth pain into Anygum's OIDC backbone.
You can balance technical accuracy with user experience. You care about whether the customer understood the answer, whether the product pain was addressed, and whether the issue will happen again.
Adoption at is measured in usage, not sentiment — 2,212 new patients and 9,650 schedules onboarded, two years seven months into the engagement and counting — because he treats 'did the fix actually land for the user' as the real question, not 'did the ticket close.' His agent work ships with evaluation harnesses (, ) that score whether an answer was actually right before it reaches anyone.
You are proactive rather than reactive. You do not wait for a perfect problem statement from the customer. You investigate, infer, clarify, and anticipate what they may need next.
Founder triage, by necessity: 's clinic-floor observation became product direction before anyone filed a request; 's federated OIDC came from anticipating client auth friction, not responding to a ticket about it. He's spent a decade investigating ambiguity because nobody was going to hand him a clean problem statement.
You go deep before escalating. You investigate as far as you reasonably can, so that when you involve engineering or product, they receive a clear problem statement, relevant evidence, and a useful recommendation.
A decade of writing the reproduction-plus-evidence-plus-recommendation escalations his own engineers received — as 's CIO and 's embedded technical lead, 'it's broken' was never an acceptable handoff, from himself or to him. That discipline is the same one behind 's CI/CD gates: nothing ships without evidence attached.
Has he actually debugged an AI agent's wrong output?
Yes — VisaPros and NewScriber both ship with evaluation harnesses that score outputs before go-live, and Global Jute's agent changes only reach production through CI/CD gates.
Debug the AI itself
When agent output is wrong, the fix is in the prompt, the retrieval, the routing, or the eval — not a canned apology.
You have genuine AI fluency at builder level. You have built agents, automations, workflows, or internal tools using AI systems such as Dust, Cursor, Claude Code, n8n, GPT-4, Claude, Mistral, Gemini, or similar platforms. You are not just good at prompting. You understand how to turn AI into operational leverage.
The JD names the exact platforms he already ships on: Claude Code (Anthropic's own Claude Code 101 and Introduction to Agent Skills certifications, earned rather than claimed) and n8n ('s entire agentic pipeline runs on n8n — cron-driven scraping, ranking, and dual-host scripting behind grounding evals). His daily development is agent-led — Claude-led, Gemini a genuine second — and multi-agent systems are his production work, not a demo track: on Google's A2A protocol, 's Claude-powered CMS behind CI/CD gates.
Formal support-stack ownership at scale — Zendesk-class tooling, queue volume, CSAT/TTR tracked as KPIs — is the honest ramp; his metrics discipline has come from instrumented pilots and evaluation harnesses, not a ticketing platform.
Weeks — the instrumentation habit already existsHe already treats every pilot as measured, not narrated ('s 2,212 patients and 9,650 schedules; evaluation harnesses scoring and before go-live). His pattern on unfamiliar systems is weeks-to-production — Google's A2A protocol and MCP each went from zero to shipped within weeks — and that's the same ramp for adopting Dust's own support-metrics stack.
He hasn't operated Dust itself in production yet — product expertise is day-one work, not a day-one fact.
Weeks — the platform is open sourceDust's core is public at dust-tt/dust, and his pattern on deep new systems is weeks-to-production. Step one is the same as it would be for any role here: rebuild his own workflows on Dust first, then arrive at customer and agent scenarios as a practitioner, not a reader of the brochure.
Professional French is in progress — working fluency is real but building.
Ongoing daily — publicly verifiableA 900+-day daily French streak, unbroken, with A2 grammar explainers taught publicly on YouTube. Dust's working language is English, and his delivery has always run in English; French adds Paris-office depth rather than gating day-one effectiveness.
Ask Fauzul's AI
'Support is not a cost center — it's one of the most important product surfaces we own.' Agreed — I've been engineering that surface for a decade, from EKAGRA's clinic floor to Global Jute's CI/CD gates.
From his recommended response · Senior AI Support Engineer (Paris)
See how it works, pattern to system.
For — a Dhaka-based export house serving global importers — Fauzul replaced the traditional CMS with an agentic one: the client describes the change they want in plain English, an agent drafts it as a pull request, and CI/CD verification gates ship it to production. The client publishes without waiting on developers. It's the closest thing to Dust's own support-deflection thesis he could have built himself — a recurring, developer-dependent request engineered into a system that runs itself, with governance designed in so the deflection stays safe to trust.
A 26-year-old export house that stopped filing developer tickets for content changes
Absorb — take the messy escalation
Investigate logs, code, product behavior and agent output directly — across nine client geographies, the ambiguous, technically complex case has been the default, not the exception.
Resolve — precision under ambiguity
Give a clear, accurate answer even when the underlying issue is unclear — 's clinic floors and 's embedded pod both demanded resolution without a clean problem statement handed to him.
Systematize — engineer it out of the queue
Turn the second occurrence into a system: 's plain-English CMS, 's n8n pipeline, evaluation harnesses that catch bad output before a customer does.
Feed back — high-context product signal
Champion the pattern back to product and engineering with reproduction, evidence and a recommendation attached — the same discipline behind 's delivery playbook, written while scaling to 22 engineers (52+ professionals total FT & contract).
Objections, answered first.
Candid answers to the hard objections — why a founder without a titled vendor support seat should own Dust's Enterprise escalations, what 'debugging the AI itself' actually looks like, and the honest state of his French.
Ask Fauzul's AI
Connected capabilities.
The stack behind the claims — multi-agent systems in production, evaluation harnesses that catch bad AI output before customers see it, and the deflection architecture (Global Jute's plain-English CMS) that is the closest thing to Dust's own support-agent thesis he could have built himself.
Agentic Development · Agentic Workflow · LLM Integration · MCP · A2A · Google ADK · Prompt Engineering · Prompt Caching · Computer Use · Eval / harness hill-climbing · Token & cost-aware context engineering · Firecrawl · Cursor · Bolt · Firebase Studio · Nano Banana · Claude Code · Claude Skills · Claude Code Subagents · Claude hooks & tools · Anthropic Claude · Google Antigravity (agy CLI) · Gemini Spark (email topic loops) · Gemini Omni (first version, via AI Studio) · On-device Gemma 4 E2B (this site chat · WebGPU + MediaPipe) · Kimi via OpenRouter (NewScriber) · Gemini TTS via Azure (NewScriber)
Multi-agent systems in production (A2A · MCP) · Claude in production (Claude-led, Gemini second) · n8n agentic workflows · RAG & retrieval · Prompt engineering & agent evals · LLM integration
Client communication · Requirements gathering · Specifications management · Enterprise problem identification · Solution architecture · Technical scoping · Delivery management · Stakeholder alignment · Executive communication
REST / GraphQL / Webhooks · Stripe · Vipps · third-party SaaS · TypeScript / Node.js · Python (AI & data) · Postgres · Docker & serverless · GitHub Actions (CI/CD)
OAuth2 · OIDC · RBAC · JWT · Auth0 · Multi-tenancy · Magic-link · Passport · HIPAA · GDPR / VAT · WCAG 2 · Auditability
React · Angular · Next.js · Astro · Redux · RxJS · Tailwind · Web Audio · WebGPU · Storybook · TanStack Query · TanStack Start
Ask Fauzul's AI
What I bring that the job description doesn't ask for.
The Aptitude / Attitude / Agency spec is the floor. Four things I'd bring into the support surface on top of it — each one bends the curve on repeat pain, not just today's queue.
Docs, internal tools and responses people actually use.
I came up as a UX Engineer, so the artifacts support leaves behind — the help doc, the internal triage tool, the phrasing of a hard answer — are things I design, not dash off. 's owners run their whole site by describing changes in plain English because I built the interface to be that legible. Deflection only works if the thing you deflect to is actually clear.
I level up the team around me, not just my own queue.
The JD asks how support onboards and reviews quality — that's a muscle. I've coached engineers into senior roles abroad, with alumni now leading at organisations in Norway (2), Canada, Germany, plus two leading local Bangladeshi tech companies, and I wrote 's onboarding and delivery rituals from scratch. A senior support hire who makes the juniors better is worth more than one who just closes hard tickets.
A support surface handling sensitive data, engineered to be safe.
Support sees everything — logs, customer data, the messy edges. I've led products into GDPR, HIPAA and WCAG 2 compliance across five client organisations and built federated OIDC and instance-level RBAC at . When an automated deflection system touches customer data, I already know where the guardrails go, because I've built them before.
I read a support failure as a business event, not a ticket.
Eight years as a co-founder and CIO means I see the run-rate cost of a recurring issue, the churn risk behind a frustrated escalation, and the expansion hiding in a well-handled one — across nine client geographies and five time zones. Support strategy is a commercial lever, and I've spent a decade pulling it from the owner's seat.
[ Support as a product surface ]
The best teams don’t just answer tickets. They engineer them away. So do I.
Intentional about Paris — a 900+-day daily French streak (public, unbroken, ongoing) and a deliberate EU/EEA relocation plan; this JD's own relocation package (up to €10k) and visa support make the landing concrete. Office-first is a preference, not a concession: the highest-leverage moments of a decade of escalations happened physically inside customer and clinic environments.
— Relocating to Paris · 994-day daily French streak, unbroken →