Applied AI Engineer, Beneficial Deployments · London, UK
Raise the floor —mission partners, public goods.
Beneficial Deployments raises the floor — Claude for nonprofits, foundations, and mission-driven teams in education, global health, economic mobility, and life sciences. Fauzul's spine is mission-adjacent delivery and enablement: UNDP Dhaka, HIPAA, GDPR/VAT, plus Claude CMS as enablement-shaped production proof (partner runs the site themselves). He wants to grow public-goods MCP/Skills packaging here. stays off this page's deep band — that connectivity story belongs to the Connectivity EM seat.
- Education
- Global health
- Economic mobility
- Life sciences
Beneficial Deployments puts Claude where it raises the floor — education, global health, economic mobility, and life sciences.
How the work maps
Four surfaces Beneficial Deployments actually needs
Not a generic enterprise architect pitch — enablement, evals, public goods, and mission constraint fluency.
Partner enablement
Pair programming, demos, tutorials, and mentoring — the Beneficial Deployments motion of making partners successful, not just shipping a one-off prototype.
Evals & harnesses
Advising on evals, hill-climbing agent harnesses, context engineering, and cost awareness inside real client delivery — until the system holds in production, not just in a demo.
Benchmarks, MCPs & Agent Skills
Turns repeating patterns into reusable skills; adds hooks and tools for cost and token efficiency — first-party fluency for packaging partner patterns into shared infrastructure.
Constrained delivery fluency
UNDP Dhaka internship, HIPAA health-tech, GDPR/VAT commerce, and WCAG across portfolio platforms — compliance-heavy delivery is familiar terrain (detail in the JD map below).
After joining
First solutions on Beneficial Deployments
Advise and enable first — then productise what worked into shared Skills, MCPs, and eval harnesses.
Pair-program with one mission partner on evals
Hill-climb a harness against their real workflow; leave golden traces and a safety/latency scorecard they can re-run without him in the room.
Draft the first public-goods packaging candidate
Turn one partner-specific Skill or MCP pattern into a shared draft — the distribution gap he named as partial, closed in public.
Publish enablement content from the engagement
Tutorial or sample code so the next nonprofit inherits the learning — teaching as delivery, not a side project.
4+ years as a Software Engineer, Forward Deployed Engineer, or technical founder.
Co-founded and grew it to 22 engineers (52+ professionals total FT & contract), distributed remote-first across deliveries to Zurich, Oslo, Dubai, Florida, Montreal, London and more. A decade as the technical founder who drops into ambiguous environments and ships — the exact FDE/founder arc the JD names.
Production experience building LLM-powered applications — prompting, context engineering, agent architectures, evaluation frameworks, and deployment at scale.
Shipped Claude into production for (full flow in Production proof below). Built agentic 0→1 systems (, , ) with context engineering, multi-agent patterns (A2A, MCP), and n8n orchestration — the same packaging habits Beneficial Deployments wants as public goods.
Builder credibility that earns trust with technical founders and engineering teams — you've shipped products and can speak from experience.
Speaks as a peer who has shipped, not a slide-deck advisor: sat with deploy and procurement in the same room for a decade. Technical founders and engineering leads get someone who has worn the hats, not someone narrating from the outside.
Experience working in ed-tech, healthcare, scientific research, nonprofit, or other mission-driven organizations — understanding their unique challenges and constraints.
IT Intern at UNDP, Dhaka, Bangladesh, Oct 2015 – Feb 2016. HIPAA-compliant EHR delivery at (3,242 patients · 13,340 schedules · tenure 2 yrs 8 mo). Also shipped GDPR + VAT commerce for 15,000+ EU users at (acquired by Loopcloud, 2026) and WCAG-aware flows across portfolio platforms. Mission-constrained delivery is familiar: limited budgets and stakeholders who need enablement more than frontier demos.
A love of teaching, mentoring, and helping others succeed.
Built 's mentoring culture across a distributed team; alumni now lead engineering teams in Norway, Canada, Germany and Bangladesh. Creates technical content and demos as a habit — the same motion Beneficial Deployments needs for partner enablement and self-service content (docs, tutorials, sample code).
Scrappy mentality — multiple hats, building from scratch, driving clarity in ambiguous situations, doing whatever it takes to further the mission.
Technical founder who wore engineering, product, GTM and delivery hats for a decade. Validates before over-building; turns fuzzy briefs into tickets and ships. Comfortable scaling team process from 1 to N because he did it at .
Serve as a deep technical partner on evals, agent architectures, context engineering, and cost optimization; provide hands-on pair programming and code contributions.
Daily practice is pair-programming-shaped: Claude Code + Claude Design in real client delivery, hill-climbing prompts and harnesses until they hold in production. He turns repeating patterns into Agent Skills and adds hooks, tools, and plugins when they cut cost, tokens, or delivery time. Context engineering and cost awareness keep production Claude changes reviewable and deployable rather than flashy demos.
Develop public-goods infrastructure — benchmarks, MCPs, and Agent Skills — that benefits entire ecosystems beyond individual partnerships.
Already packages reusable Claude Skills and MCP-connected tooling from common workflow patterns in his own repos — first-party Anthropic fluency, not slide-deck familiarity. Has not yet published Anthropic-scale public benchmarks or ecosystem MCP packs as open public goods.
AdaptThe instinct and first-party fluency are in place. On Beneficial Deployments he would turn partner-specific harness patterns into shared Agent Skills, MCP servers, and eval benchmarks — the same playbook he used to codify delivery habits into reusable team practice — aimed at raising the floor for every mission partner, not just the one in the room.
Formal, metric-driven evaluation frameworks as a first-class product surface (advising partners on evals / hill-climbing harnesses).
Builds technical evals and POCs to decide what ships, and brings strong automated-testing and compliance-auditability discipline — but has not yet stood up a formal partner-facing agent-eval harness as a public product surface.
AdaptTreat partner eval quality the way he treats compliance: designed in, measurable, auditable. Formalise golden traces and accuracy/safety/latency scorecards into reusable harnesses partners can run themselves — a 4–8 week ramp on habits he already has, exactly aligned with Beneficial Deployments' advising-on-evals mandate.
Top strength
Claude already in production — Anthropic-certified to package it
Claude already runs in a client production stack — see Production proof. Verify the first-party Anthropic credentials below: the MCP and Skills fluency Beneficial Deployments wants to scale as public goods.
Also day-one
Supporting strengths
Mission-adjacent scar tissue, not tourism
UNDP Dhaka internship, HIPAA health-tech, GDPR/VAT commerce — constrained partner context without tourism. Numbers and platforms live in the mission JD row above.
Teacher-operator who scales from 1 to N
Codified delivery playbooks so impact outlived any single engagement. Built for the enablement + public-goods half of this role, not only the one-off prototype (mentoring geography detail is in the teaching JD row).
Safety-first builder temperament
Chose Claude for client delivery because Anthropic's steerable, reliable posture is what stands up in production. Treats compliance and predictability as product surfaces — the same bar mission partners need before they trust AI with education, health, or economic-mobility workloads.
Adaptable gaps
Named openly, with a bridge
Anthropic-scale public-goods releases (shared benchmarks, ecosystem MCP packs, published Agent Skills) beyond personal/repo-level packaging.
First-party Agent Skills fluency plus a habit of turning engagement patterns into reusable playbooks. On the team he would productise partner harnesses into shared Skills/MCPs/benchmarks so the next nonprofit inherits the learning.
Formal partner-facing eval harnesses as a first-class surface.
Extend existing eval/POC and test discipline into golden sets and safety/latency scorecards partners can run — designed-in measurement, not trial-and-error demos.
Navigating a fast-scaling frontier lab versus running his own consultancy; hybrid London office expectation (~25%).
A decade leading distributed teams and reporting cleanly into client hierarchies makes Anthropic's structure familiar. Relocating to London deliberately; commits to local business hours and the office hybrid bar from day one, with visa sponsorship required.
Production proof
Claude in production — built for partners who need the floor raised
I chose Claude for client delivery because mission-driven partners need predictability, not demo theatre — and enablement-shaped work taught me to leave a partner able to run the system themselves, not lean on me.
Fauzul · Beneficial Deployments · enablement-shaped Claude
Global Jute — Claude CMS as partner enablement
For — a 26-year-old Dhaka export house — a Claude-based system lets the partner express the change in plain natural language. Claude turns that request into a pull request and triggers CI/CD to production. Enablement baked into the product: the partner runs the site themselves. That is Beneficial Deployments' deep band — not Connectivity's platform story, and not Zürich Industries' Sales-paired POC framing of the same artifact.
- Natural-language intent → Claude-authored PR → automated CI/CD deploy — every change reviewable and revertible in Git.
- The client manages their own marketing drives without a developer middleman — the Beneficial Deployments enablement shape in miniature.
- Proof that frontier AI ships into a specification-sensitive, resource-constrained business — not a demo.
Enablement flow · Global Jute Trading
- 01IntentPartner describes the change in plain English
- 02ClaudeAuthors a reviewable pull request
- 03ReviewHuman gate before merge — nothing ships unreviewed
- 04OwnUpdates go live same-day, unblocked by a dev queue
- Bolt.new — the $700 nightAll-night agentic curiosity — the scrappy energy mission partners need when budgets are tight.
- Cursor · Gemini · broader ecosystemWent deep across the tooling landscape — every tool opened my eyes to where software was heading.
- Claude Code + Claude Design + Agent SkillsThe exception. Brought into real client delivery — packaging Skills and MCP patterns from production work. The reason Beneficial Deployments is the seat I want.
How agents get run
How I actually run agents
Claude-led for client delivery. Everything else is deliberate experiment: skills, hooks, harnesses, and loops that cut cost, tokens, and time-to-ship.
Skills from repeating patterns
When a workflow shows up more than once, I turn it into an agent skill with reusable instructions and triggers, rather than a one-off prompt.
Hooks, tools, and plugins for cost and speed
Most builds add hooks, tools, or plugins aimed at lower spend, tighter token use, and faster delivery. It is deterministic control in the places where prompts alone waste budget.
Harness and model fit per task
Claude Code for primary client build work; Antigravity and others when the task or token budget asks for it. This site's chat runs on-device Gemma 4 E2B (WebGPU + MediaPipe, model hosted on R2). NewScriber scripts with Kimi via OpenRouter and voices with Gemini TTS via Azure.
Technical inventory
The builds that prove the seat — Claude already in a client production stack, Agent Skills and MCP packaging habits, regulated delivery across health and commerce, and the teaching instinct that turns one partnership into ecosystem leverage.
Applied AI & Agents
- Claude-led daily development (Gemini second)
- Production LLM apps (prompting · context · agents)
- MCP & Agent Skills packaging
- Eval / harness hill-climbing
- n8n agentic orchestration
- A2A multi-agent patterns
Partner enablement
- Pair programming & technical advising
- Demos, tutorials & sample code
- Mentoring (Norway · Canada · Germany · Bangladesh)
- Client-facing delivery leadership
- Ambiguous briefs → shipped systems
Mission & compliance
- UNDP Dhaka (IT Intern)
- HIPAA (EKAGRA · 3,242 patients · 13,340 schedules)
- GDPR / VAT (Jamahook · Zürich · 6+ yrs ELO · Loopcloud 2026)
- WCAG 2
- Compliance as a product surface
Frontend
- React
- Next.js
- TypeScript
- Tailwind
- TanStack Query
- Design systems
Backend
- Node.js
- NestJS
- Python
- Go
- Postgres
- REST / GraphQL
Analogous builds
Projects that map to this seat
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.
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.
Submitted as the capstone for the Google × Kaggle 5-Day AI Agents Intensive (2026). VisaPros is an agentic visa eligibility advisor utilizing a Google ADK agent-to-agent (A2A) hub-and-spoke architecture — parsing documents once and running six parallel country agents as FastAPI microservices to evaluate fit across destinations.
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.
Difficult questions
Honest answers a Beneficial Deployments panel should ask — why this mission seat over a pure enterprise architect role, how he teaches without condescending, and how he holds the safety bar while enabling partners who are early in their AI journey.
This role is teaching, mentoring, and enablement for partners early in their AI journey — not frontier model development. Is that really what you want?
Yes — and I would rather you hear that from me than from a candidate who treats enablement as a consolation prize. The highest-leverage applied work I have done has always been making someone else successful: mentoring engineers into leadership, turning fuzzy client briefs into shipped systems, and lately putting Claude into a client's hands so they run their own site without a developer middleman. Beneficial Deployments is that motion at Anthropic's mission altitude. I am not applying to escape hard engineering; I am applying to put hard engineering in service of partners who need the floor raised.
You run your own consultancy — why leave for Beneficial Deployments specifically, rather than another Applied AI Architect seat?
Because the mission match is not abstract. UNDP Dhaka, health-tech under HIPAA, EU commerce under GDPR, and a decade of constrained, compliance-heavy delivery taught me that the partners who need AI most often have the least spare engineering capacity. Beneficial Deployments is the team built for that reality: advise on evals, hill-climb harnesses, pair-program, and leave behind public goods. The Architect seats are about enterprise choreography; this seat is about raising the floor. That is the chapter I want.
How do you keep Claude safe and reliable enough for education, health, or nonprofit partners who cannot absorb a bad failure mode?
I treat predictability as the product. Compliance and evaluation are not back-office tax — they are how you earn the right to deploy. At that meant HIPAA as a shipped surface; with Claude CMS it means every generated change ships through the same pull-request gate as human-written code. For mission partners I would start with eval harnesses, narrow tool access, and clear escalation paths before any ambitious agent loop — and feed partner-specific failure modes back to product and research, exactly as the JD describes.
You marked public-goods MCP/Skills infrastructure as partial. Why should we trust you to build ecosystem tooling?
Because I already package Skills and MCP-connected workflows in practice, and I hold first-party Anthropic credentials for that stack — I am not learning the nouns on day one. What I have not done is publish them as Anthropic-scale public goods. The gap is distribution and productisation, not fluency. Beneficial Deployments is where partner-specific harnesses become shared benchmarks and Agent Skills; that translation from one engagement into team practice is how I scaled 's delivery playbook.
Will you relocate to London and meet the ~25% office hybrid expectation? Do you need visa sponsorship?
Yes on both counts I can control: London is where I want to base this chapter — my sister lives there, I have delivered for London clients for years, and I commit to local hours and the office hybrid bar from day one. I will need visa sponsorship; if you make an offer I will work the process in good faith with your immigration support.
Beyond the checklist
What Beneficial Deployments gets past the JD
Mission scar tissue and enablement instinct — the Beneficial Deployments altitude past the checklist.
UNDP Dhaka as origin scar tissue
IT Intern at UNDP Dhaka (Oct 2015 – Feb 2016) — early exposure to mission-constrained delivery where spare engineering capacity is rare. The Beneficial Deployments altitude started here.
Jamahook GDPR/VAT + EKAGRA HIPAA
EU commerce under GDPR/VAT (, acquired by Loopcloud 2026) and HIPAA clinical delivery at — regulatory constraints designed in from day one, not bolted on after an audit finding.
Teaching as a delivery surface
Alumni now lead engineering teams in Norway, Canada, Germany, and Bangladesh — proof the teaching sticks after he's out of the room, the exact posture a partner who is early in their AI journey needs.
Start a conversation
Intentional about London, UK — hybrid office-ready (~25%), visa sponsorship required. Ready to advise on evals, pair-program with mission partners, and help shape the public goods that raise the floor.