Applied AI Engineer - Agentic Workflows — built to scale reasoning and workflows.
Fauzul brings a decade of full-stack production engineering (React, Go, and 7+ years TypeScript) combined with frontier agentic AI development. He is exactly the engineer Cohere is looking for: someone who has built and shipped multi-agent systems in production (Claude-powered CMS at ), capstone frameworks (Google × Kaggle 5-Day Agent Intensive using a Google ADK A2A network), and autonomous pipelines ( using Go/n8n/Gemini TTS). His background scaling distributed teams of 22 and collaborating with external CTOs maps directly to your Stakeholder Mastery and Mentorship criteria. He treats reliability, observability, safety, and rigorous evaluation (measuring agent accuracy, safety, latency) as core product surfaces.
Full-stack production engineer + agentic builder
Applied AI Engineer - Agentic Workflows is where my full-stack TypeScript craft and agentic AI architectures compile together. I build agents that plan, act, and evaluate—designed for enterprise-grade trust.
A decade of production engineering + Google ADK agentic systems
At , Fauzul built and led a distributed engineering team from 0 to 22 across five time zones while personally architecting core framework substrates and component libraries. He brings 7+ years of production TypeScript experience alongside frontier agentic development: (a Google × Kaggle capstone utilizing a Google ADK A2A network), (an agentic news editor and podcast pipeline (localized TTS from any source, tested with Norwegian/Bengali) (localized TTS from any source, tested with Norwegian/Bengali)), and a Claude-powered CMS running in a client's production stack. He focuses on rigorous evaluation frameworks, enterprise compliance (HIPAA, GDPR, OIDC RBAC), and stakeholder mastery.
TypeScript: 7+ years of production-grade typed, testable, and scalable full-stack software.
Google ADK A2A: — six parallel FastAPI country agents in a collaborative network.
Agentic Scraper: — multi-agent editorial pipeline with Firecrawl and Gemini TTS.
Enterprise Compliance: HIPAA ( EHR), GDPR ( checkout), federated OIDC/RBAC ().
Stakeholder Mastery: a decade translating ambiguous enterprise briefs into shipped architectures.
Production-grade TypeScript & Go
7+ years of full-stack TypeScript (React, NestJS) and Go backends in production, backed by automated testing and CI/CD.
Google ADK & A2A agentic architectures
VisaPros capstone utilizing Google ADK and agent-to-agent microservices, NewScriber Go/n8n pipelines.
Enterprise compliance & stakeholder trust
Translating ambiguous requirements, leading discussions with client CTOs, and shipping HIPAA/GDPR/OIDC systems.
Point by point
The role is Applied AI Engineer - Agentic Workflows, building production-grade agents for enterprise customers at scale. Here is how a decade of full-stack engineering, production-grade agentic architectures, and stakeholder collaboration maps onto each criterion.
Production Engineering: Substantial experience building, shipping, and maintaining production-grade software (7+ years TypeScript). You understand how to design testable, observable, and scalable code.
Started in 2015 as a UX Engineer — writes production React, Next.js, and Node/TypeScript backends daily. 7+ years of professional TypeScript delivery under (ERP ledger, billing gateways, core platforms) and client systems. Practices strict type-safety, automated testing (Jest, Cypress, Playwright), CI/CD gating, and structured observability.
Agentic Architectures: Hands-on experience building agents that plan and execute multi-step reason (A2A, Plan-and-Execute) and integrate with systems and API/tools.
Designed and built (agentic news scrape, editor, and dual-host dialogue scripting via Azure OpenAI ReAct loops) and (collaborative A2A country microservices mesh built using Google ADK). Integrates agents with external tool APIs, Postgres, and Firecrawl scraping workflows.
The LLM Stack: Deep familiarity with frontier models (chat, function calling, RAG, vector databases (Pinecone, Pgvector), etc.), and orchestration frameworks (LangGraph, CrewAI, or custom state machines).
Daily active development with Claude & Gemini APIs, function calling, RAG pipelines, and vector databases (Pgvector on Postgres, Firestore). Orchestrated agents using custom state machines and n8n graphs. Understands CrewAI and LangGraph conceptual boundaries (sequential vs. network execution, memory boundaries).
Rigorous Evaluation: Proven ability to move beyond 'feels like work' by building robust evaluation frameworks to measure agent accuracy, safety, and latency.
Built custom evaluation loops in (evaluating dual-host dialogue scripts against scraped source bodies before voice synthesis) and (evaluating visa category recommendations against grounding regulations via Google Search). Focused on metrics that gate production deployment based on correctness and safety, rather than subjective 'feels like' outputs.
Stakeholder Mastery: Experience leading technical discussions with enterprise customers to translate ambiguous business needs into concrete technical specs.
Co-founded and spent a decade operating as the primary technical interface for enterprise accounts in London, Zurich, Oslo, Montreal, and Dubai. Translates fuzzy briefs into user stories, design requirements, and API specs. Partnered with client CTOs to scope and coordinate deployments.
Experience mentoring distributed teams and setting the architectural standards for agentic systems.
Grew 's engineering pod from 0 to 22 distributed engineers. Mentored developers into senior roles internationally (Norway, Canada, Germany) and set the architectural standards for auth backbones (OIDC/RBAC ), double-entry SaaS (), and multi-agent meshes.
Strengths & honest gaps
7+ years of production-grade TypeScript & full-stack craft
Substantial experience shipping typed, testable, and scalable full-stack software in production for enterprise clients. Fluently bridges React/Zustand frontends, Go/FastAPI backends, and SQL schemas.
Google ADK & A2A agentic architectures in production
Proven capabilities building agent-to-agent meshes ( utilizing Google ADK), multi-agent editorial engines (), and live natural-language CMS integrations ( with Claude).
Rigorous evaluation & observability mindset
Bakes in automated verification loops, measuring accuracy, latency, and correctness. Replaces subjective agent evaluations with structured gating rules and search grounding.
Enterprise stakeholder mastery & client-facing trust
A decade of experience negotiating and executing complex contracts, defining technical specifications directly with clients, and deploying HIPAA-compliant ( Health) and GDPR-compliant () systems.
People leadership & team mentorship
Built, scaled, and managed a distributed team of 22 engineers at remote-first, coaching alumni who now lead teams in Norway, Canada, and Germany.
No material gaps for this role.
Candid answers
Candid answers — from balancing rapid iteration with enterprise quality standards to the technical details of his A2A, RAG, and evaluation setups. Ask the agent anything else.
You're a co-founder and CIO — why apply for an Applied AI Engineer role?
As 's CIO, I co-led a distributed team of 22 and handled the client-facing solutions. I loved the technical scoping, the customer integration, and the agentic engineering—but the agency overhead (billing, legal, admin) grew around it. This role gets me back to 100% focus on shipping agentic workflows and solving customer problems at scale, with Cohere's models behind it. That's not a step back; it's focus.
The role involves translating ambiguous client needs. Can you share an example of how you've handled this?
For and , the brief was simply 'we need a unified identity gateway for our SaaS suite.' I led the technical discussions with client CTOs, mapped out the OIDC federated login flow, defined the instance-level RBAC schema in PostgreSQL, and designed the multi-tenant isolation rules. The resulting technical spec was implemented by a pod of 3 and became 's central backbone.
Immigration rules drift constantly. How does VisaPros solve for this?
It doesn't rely on model training weights. Each country agent (UK, CA, US, etc.) uses real-time grounding via Google Search to fetch the latest official PDF policy documents and regulatory changes from government sites before performing the evaluation. This guarantees the recommendations are accurate and legally sound.
What's your approach to evaluating agent behavior?
I treat agents like microservices—they must have typed input/output contracts, structured logs, and automated tests. For evaluations, I build a separate evaluator agent (using a different model or search grounding) to verify correctness and safety, and track execution latency. If the evaluator flags a response, the script is rejected and logged for debug.
The stack & the builds
The builds that prove the capabilities — agentic AI in production, full-stack TypeScript/React and Go architectures, and the compliance layers (HIPAA EHR at EKAGRA, GDPR VAT at Jamahook, OIDC/RBAC at Anygum) that make enterprise customers trust agentic pipelines with sensitive data.
Editorial export site for Global Jute Trading Ltd — a Dhaka-based sourcing & exporting house with 26 years as a company and 35 years of jute trading under its principals. Built to support their push into deeper international markets, with a particular focus on Canada.
Agentic news scraper & editorial audio briefing network. Scrapes tech and business news via Firecrawl, curates & scripts dual-host dialogues with Kimi via OpenRouter, and renders high-fidelity multilingual voices via Gemini TTS on Azure.
A new LinkedIn for orgs that actually deal. TagRamp connects organizations from goals to meaningful business deals through AI-mediated intros. Built solo during the World's Largest Hackathon (May 2025) by Bolt.new and submitted to Y Combinator; now scaling with active enterprise pilots and CXO-level partnership discussions.
Strategic health-tech partnership and equity investment in Bangladesh's most advanced clinical practice for wound care, diabetes, and nephrology. Spearheaded high-utility, HIPAA-compliant EHR clinical dashboards and scaled digital patient advocacy.
Anygum is a working marketplace-style platform where anyone can host their own API and builders can compose bespoke apps that pull services from multiple vendors. Identity runs through Cloudsight — custom auth on Auth0 — so provider and consumer context share single sign-on, RBAC and org controls, multi-instance of the same app for different teams, easy telemetry, and a single payment funnel for consumption. Ideation started as an internal auth problem for SaaS distribution; it was spun out as Anygum. Elobooks.net uses it. Distribution limited wider market reach; the product itself works.
Jamahook AG is a Zürich-based music-tech company and a direct ELO client for over 6 years — the world's first AI-powered music matching ecosystem, launched in October 2020, with a desktop agent, VST/AU/AAX plugins, and a web portal. Producers find complementary material through acoustic and psychoacoustic audio annotation and analysis, on an algorithm developed in collaboration with Fraunhofer IDMT — the institute behind MP3. I created the first web product and admin panel from a blank canvas, then stood up the full-stack team (React + Node/NestJS on an NX monorepo) and stayed hands-on through the first market release. The dedicated team I helped build still ships modern integrations; since 2023 my seat has been primarily client and business relationship. Loopcloud acquired Jamahook in 2026, and the matching engine now powers Loopcloud's AI sound matching.
Ready to put AI to work?
I want to relocate — Toronto is always my first choice, London a strong second, and I'm open across Europe and North America too. Toronto is Cohere's co-HQ: I've lived six months in and around the city and built real professional and personal ties with clients there and in Montreal. London is a natural European hub: my sister lives there and my best friend is in Norway. Either base puts me inside Cohere's core orbit.