Skip to content
Currently available — for the right work·France903+ Day French Streak·2026 Q2 calendar — open now
Application · Founding Full-Stack Engineer (Back-End Leaning) · London

Patent drafting is a document workflow with a human who has to sign it. I have already shipped one of those.

Founding full-stack engineer, London. Nine years co-founding and building ELO — a multi-tenant RBAC backbone, a HIPAA EHR live with 64 doctors, a checkout serving 15,000+ users to acquisition — and a production workflow where a model drafts, a human approves the diff, and CI/CD ships it.

12 of 14 claim elements mappedRamps stated, not argued around
Read the claim chart
Claim chart · Solve Intelligence14 elements
Cl.Element
1Expert in Python
2Experience with databases (Postgres)
3Familiarity with front-end and back-end development and full-stack frameworks
4Build the core, scalable foundations of the backend for a huge market, from the ground up
9 yrs
Co-founding and building ELO
UX Engineer → Principal Staff SWE → CIO, never off the codebase
22
Engineers hired, onboarded, led
52+ FT and contract; remote-first from 2020
15K+
Users on shipped commercial software
Jamahook — GDPR and EU VAT in the payment flow, acquired 2026
5
Client orgs led into compliance
GDPR · HIPAA · WCAG 2
Specification

The enabling disclosure

The proof Solve should actually weigh: a production workflow where a model drafts and a human approves before anything ships.

[0001]

Founding-team range, already lived

Co-founded in 2017 and still writes the code. Grew it to a 22-person product engineering venture firm (52+ FT and contract), hired and onboarded the team, ran it distributed remote-first from 2020, and moved from UX Engineer to Principal Staff SWE to Engineering Manager to CIO without leaving the codebase. Solve is asking someone to do the engineering-team half of this alongside three founders — he has done both halves at once.

[0002]

Agentic orchestration from higher ground

He builds agentically now — Claude-led daily as the primary environment, with first-party Anthropic grounding (Claude Code 101, Introduction to Agent Skills), custom MCP servers, and multi-agent systems running in production. That moves where the bottleneck sits: the scarce input stopped being how fast someone types idiomatic code and became knowing what to build, where the irreversible mistakes live, and which guardrails must exist before a line is written. Nine years of production judgment supplies those; the agent layer executes inside them. It is why a new language or a new library surface is a matter of weeks rather than a career chapter. is the sharpest demonstration: eight days from scaffold to a working end-to-end system in which investor and founder agents negotiate to an explainable number, a human stays in the loop on the consequential calls, and every outcome seals into a tamper-proof hashed record. It runs Claude (Anthropic SDK) with Gemini alongside, and ships a deterministic rule-based fallback, so the product still runs correctly with no model key at all — the model is an accelerant, never a dependency.

[0003]

The document workflow, not just the model

For he replaced a traditional CMS with a Claude-driven workflow: the client team writes the change in plain English, it lands as a reviewed pull request, and CI/CD handles the preview build and deployment. Nothing a model generates reaches production without a human approving the diff — which is precisely the trust contract between an AI drafting tool and the patent attorney whose name goes on the filing.

[0004]

Confidentiality designed in at the schema

's identity backbone isolates every consumer organisation on its own instance under RBAC multi-tenancy on Auth0 with a layered proprietary authorization model — tenant separation as a structural decision, not a WHERE clause. runs a HIPAA-compliant EHR live with 64 doctors across 7 centers; automated GDPR and EU VAT inside payment flows. Five client organisations were led into GDPR, HIPAA and WCAG 2 compliance.

[0005]

The engineer who sits in the customer meeting

Nine years of discovery run directly with clients across seven countries, plus the weekly and monthly stakeholder syncs on 's financing build — solving requirement problems before they reached the codebase. CSPO-certified, with an IIBA product-management foundation. Extracting the roadmap from a customer conversation is an existing habit, not a new responsibility.

Claim chart

Every line of the posting, mapped to evidence

Every line of the posting — must-haves, preferred, nice-to-haves and the responsibilities — mapped to evidence, with the Python verdict stated honestly rather than argued around.

Cl.Claim elementMapped evidenceStatus
1Must haveExpert in PythonPython is a shipping language here, and how it ships is the point: six parallel FastAPI agents in on Gemini 2.5 Flash Lite with Google Search grounding, custom MCP servers, ingestion and ETL, backed by a Kaggle Python for Data Science certification. It is built the way he builds everything now — orchestrating agentically from the architecture downward, Claude-led daily, rather than typing every line. Nine years of production judgment sets the guardrails the agent layer executes inside: typing discipline, migration safety, test gates, review before merge. The commercial scar tissue is TypeScript/Node and Go; the operating altitude transfers whole.Ramp: house idiom in a large Python codebase — async conventions, ORM patterns, the local test culture. That is weeks of absorption, not a rebuild, and it is the layer agentic orchestration closes fastest. Knowing which guardrails must exist before a line is written is the part that took nine years, and it arrives on day one.Ramping
2Must haveExperience with databases (Postgres)PostgreSQL is the default store across the portfolio and the schemas are his own. 's identity service isolates every consumer organisation at the schema under RBAC multi-tenancy on Auth0 with a layered proprietary authorization model; 's HIPAA EHR carries real clinical load (64 doctors · 3,242 patients · 13,340 schedules across 7 centers); 's checkout served 15,000+ users; 's ingestion and ETL platform models 47,091 matches across 1,207 players. pgvector for embedding work. And it is current rather than historical: runs Prisma on PostgreSQL across 11 models, with the ledger's integrity rules enforced in the schema itself — balances derived from their legs rather than stored, and no UPDATE or DELETE path at all, so corrections are compensating entries and history cannot be quietly rewritten.Mapped
3Must haveFamiliarity with front-end and back-end development and full-stack frameworksNestJS in production (, ), TanStack Start (, ), Next.js and Astro across client builds, Go services behind Cloud Run. The front-end/back-end boundary has never been someone else's job — he started as 's first UX Engineer and reached Principal Staff SWE without ever leaving either side of it.Mapped
4ResponsibilityBuild the core, scalable foundations of the backend for a huge market, from the ground up is the closest match in his history: conceived and architected from an empty repo into a multi-tenant marketplace where providers register once and every consumer organisation runs on an isolated instance. — the ERP built end to end for multi-organisation business operations — runs on that backbone. 's ingestion, ETL and analysis platform was likewise taken from nothing to production single-handed.Mapped
5ResponsibilityBuild state-of-the-art AI systems with a clean, intuitive, human-centric user interfaceThe pairing is the whole profile: he designs the interface and writes the agent layer behind it. 's CMS puts a non-technical export team in front of Claude without ever exposing the model; 's ordering platform (React + Zustand on a Go service, on-demand Cloud Run) does the same for restaurant operators. Figma-to-production discipline with a Storybook design system on .Mapped
6ResponsibilityExtract insights from customer meetings to determine what we need to buildNine years of running discovery directly with clients in London, Zurich, Oslo, Dubai, Florida, Montreal and Laval — including the weekly and monthly stakeholder syncs on 's financing build, where the job was to solve requirement problems before they reached the codebase. CSPO-certified; the roadmap conversation has never been someone else's handoff.Mapped
7ResponsibilityPartner with the founding team to build, scale, and set the culture of the engineering teamThis exact job, from the other chair. Grew to a 22-person product engineering venture firm (52+ professionals FT and contract), hired and onboarded it, moved it remote-first in 2020 (+50% productivity, +100% satisfaction), and coached engineers out into their next roles — two alumni now lead teams in Norway, plus alumni at two leading local tech firms. Culture was set deliberately, not inherited.Mapped
8About youMotivated to ship product fast: an IDE update broke a team's WSL workflow overnight, root-caused and a scoped MIT tool released the same week. went from nothing to a Y Combinator submission during the World's Largest Hackathon. was built inside a 7-day Milan AI Week sprint and now publishes to Spotify daily, in production. Most recently : an autonomous agent-to-agent deal-clearing prototype for pre-seed venture, taken from scaffold to a working end-to-end product in eight days and submitted to Y Combinator for the Fall 2026 batch. It runs as a working simulation — no real capital moves — but the whole loop is there: agents negotiate, a sandboxed audit grades the candidate codebase, and every movement of value is written as a balanced, hash-chained, append-only ledger entry — integer minor units, a single writer module, and unbalanced transactions rejected whole rather than written and reconciled later.Mapped
9PreferredExperience with Web Development (TypeScript / React)The default stack rather than a preference: React 19 + TypeScript across every active build — , , , , and fauzul.com itself. Nine years back through Angular and RxJS enterprise work (certified by the Angular team at Google), so the React choice is an informed one, not the only one he knows.Mapped
10PreferredBuilding and scaling commercial software reached 15,000+ active users with GDPR and EU VAT compliance automated inside the payment flows, and was acquired by Loopcloud in 2026 after 6+ years as an client. is live with working clinicians rather than piloted. itself is nine years of commercial delivery — software that had to earn revenue, not demo well.Mapped
11Nice to haveWYSIWYG editor experienceHe has already built the hard half of this from scratch. 's browser-native collaborative editing surface needed a document model, selection state and conflict handling written by hand over custom WebSocket sync — no library to lean on. 's publishing workflow then solved the part most editor integrations get wrong: keeping a human review gate between generated content and the live artifact. On the named libraries, precisely: TinaCMS a few years ago — real, if light — and Vercel's multiplayer block editor evaluated rather than shipped. He will not claim TinyMCE or CKEditor 5 he has not run.Ramp: a documented plugin architecture layered over concepts already implemented by hand. Editor APIs are the surface agentic orchestration closes fastest — public docs, conventional patterns, and the scarce judgment (what belongs in the editor versus behind the review gate) already settled. Two to three weeks to productive.Ramping
12Nice to haveAWS DevOpsLambda (including container images, as an early adopter), ECS, EC2 with deliberate region and cost strategy, EventBridge, SQS/SNS, CloudFormation with SAML-based provisioning, IAM and WAF. GitHub Actions pipelines replaced manual deploys across 's portfolio for a 75% cut in time-to-market; HashiCorp Vault for secrets and key management.Mapped
13Nice to haveIT security awarenessSecurity as architecture rather than a checklist: OAuth2 and SSO with multi-tenant RBAC and per-organisation instance isolation on ; HIPAA on ; GDPR and EU VAT on ; products led into GDPR, HIPAA and WCAG 2 compliance across 5 client organisations. Directly relevant to a product whose customers' entire value sits in unpublished invention disclosures.Mapped
14Nice to haveAI experienceClaude-led daily as the primary build environment — Claude Code 101 and Introduction to Agent Skills, both issued by Anthropic — with Gemini second. Multi-agent systems in production: ships daily, runs six agents in parallel, and 's AI layer was rearchitected mid-flight to backend-driven orchestration once enterprise security review demanded it. MCP servers, RAG, and evaluation harnesses for accuracy, safety and latency.Mapped
Confidentiality

The part of your product that cannot be wrong

Solve's customers hand over unpublished invention disclosures. Here is how Fauzul has already built for that standard — architecture first, checklist second.

Tenant isolation at the schema

's identity service gives every consumer organisation its own isolated instance under RBAC multi-tenancy on Auth0 with a layered proprietary authorization model. Separation is structural, not a WHERE clause someone can forget.

Compliance designed in, not retrofitted

Five client organisations led into GDPR, HIPAA and WCAG 2 compliance — including a live EHR and EU VAT correctness inside a payment flow. Compliance as a user-facing surface rather than back-office tax.

Secrets and access, handled

HashiCorp Vault for secrets and key management, AWS IAM and WAF, CloudFormation with SAML-based provisioning, SSO and JWT across multi-organisation deployments.

Nothing generated ships unreviewed

The discipline, already in production: model output lands as a pull request a human approves before it reaches the live artifact. The same trust contract an attorney needs before their name goes on a filing.

Office action

The one objection worth raising, answered

Examiner · rejection

Must have — Expert in Python. Applicant's nine years of commercial depth are in TypeScript/Node and Go. Python appears in agent and data work rather than as the primary production language.

Applicant's response

Acknowledged, and worth answering with the actual working model rather than a promise. Python ships today — six parallel FastAPI agents in , custom MCP servers, ingestion and ETL, a Kaggle certification — so neither the language nor its ecosystem is new. More to the point, it ships the way everything ships here: orchestrated agentically from the architecture downward, Claude-led daily, with first-party Anthropic grounding behind it. That changes where the bottleneck sits. The scarce input is no longer how fast someone types idiomatic Python; it is knowing what to build, where the irreversible mistakes live, and which guardrails have to exist before a line is written — typing discipline, migration safety, test gates, review before merge. Nine years of production judgment supplies those on day one, and the agent layer executes inside them at a pace hand-writing does not reach. House idiom in a mature codebase is weeks of absorption while already contributing. The altitude is what a founding engineer is hired for, and it transfers whole.

Cited references

The closest thing to your product already running in production

[R1]

Global Jute Trading — the CMS he deliberately did not build

A 26-year-old export house needed to publish spec-grade product content without waiting on a developer queue. Instead of a traditional CMS, Fauzul shipped a workflow with Claude as the harness: the client team describes the change in plain English, an agent turns it into a pull request against the real codebase, CI/CD builds a preview, and a human approves the diff before anything deploys to Cloudflare Workers.

  • The non-technical team got its own release lane; developers got their focus time back.
  • Generated content is never trusted straight to production — the review gate is the product, not a setting.
  • MDX content pipeline with remark-frontmatter, edge SSR, and a responsive image pipeline behind it.
  • Live for a real business with revenue attached, not a demo.
Claude Code
Primary build environment. Claude Code 101 and Introduction to Agent Skills, both issued by Anthropic.
MCP
Custom servers in Python and Go feeding tool state to agents without leaking secrets.
FastAPI
Six parallel agents in , on Gemini with Google Search grounding.
n8n
DAG orchestration behind 's daily multi-agent publishing pipeline.
GitHub Actions
The CI/CD lane that turns an approved diff into a deploy — 75% cut in time-to-market at .
Disclosure docket

The questions a sharp founding team would actually ask

The objections a sharp founding team would raise on a first call, including the Python must-have and the founder-taking-an-IC-seat question, answered directly.

Take the evidence at exactly its weight. Python is shipped: runs six parallel FastAPI agents on Gemini with Google Search grounding, MCP servers are written in it, ingestion and ETL run through it, backed by a Kaggle Python certification. What nine years of production incidents taught him happened in TypeScript/Node and Go — that is true and worth saying. But it matters less than it would have five years ago, because he no longer builds by typing every line. He orchestrates agentically from the architecture downward, and the guardrails that orchestration runs inside — typing discipline, migration safety, test gates, review before merge — are exactly what nine years bought. House idiom in your codebase is a few weeks of absorption while contributing. Judgment about what to build, and where the irreversible mistakes live, does not ramp at all.

Because the title he wants back is the one he never actually gave up. Across nine years at he moved UX Engineer → Principal SWE → Principal Staff SWE → Engineering Manager → CIO, and the constant through all of it was writing the software. The management chapters were things the company needed, and he did them well — grew the team to 22, hired and onboarded it, coached people into leadership abroad. But the work he chooses when nobody assigns it is building: in a week, to a YC submission, shipping daily. Joining a 50-person profitable company with three technical founders and a clear thesis is not a demotion; it is a chance to build at a velocity a services firm structurally cannot reach.

The domain is genuinely new and he will not pretend otherwise. What is not new is the pattern of getting productive inside a high-consequence domain he did not start in — a HIPAA EHR now live with 64 doctors across 7 centers, GDPR and EU VAT correctness inside a payment flow serving 15,000+ users, WCAG 2 compliance across five client organisations. In each case the discipline was the same: learn where the irreversible mistakes live, build the guardrails there first, and keep a human in the loop on anything consequential. Solve has already solved the knowledge half structurally by putting attorneys next to engineers — the useful question is whether he can build correctly against expert guidance, and that is nine years of the answer.

He has built the hard half of an editor without a library. 's real-time collaborative editing surface needed a document model, selection state and conflict handling written from scratch over custom WebSocket sync — that is the part people underestimate. A documented plugin API on top of solved concepts is a reading task with agentic orchestration behind it, measured in weeks. And the genuinely hard problem in an AI drafting product is not the editor widget: it is the trust contract around it — what the model may touch, what a human must approve, how a reviewer sees exactly what changed. That contract is already running in production in the workflow, which is the piece that usually has to be learned the expensive way.

Remote-first since 2020 across six time zones, so full UK-hours overlap holds from week one regardless of where the paperwork stands — that part is not a promise, it is nine years of operating pattern. The relocation itself is settled ground rather than intent: his sister already lives in London, so housing and landing logistics are answered before the visa conversation starts. London is also a deliberate choice rather than a tolerated condition — it is Europe's centre of gravity for AI, and the density of serious teams there is exactly the pressure he wants on his work.

The honest risk is over-processing. A venture firm delivering for external clients builds rituals that protect against ambiguity it cannot control — discovery documents, sign-off gates, written requirement trails — and a profitable product company shipping to its own roadmap needs far fewer of them. He is aware of the reflex and would deliberately run without those gates until something actually breaks. The counterweight is what the services background does give you: nine years of shipping under someone else's deadline with revenue attached, which is closer to a founding-engineer posture than most product-company résumés of the same length.

Beyond the posting

What you also get, that the requirements list does not ask for

A founding engineer is hired for the checklist and kept for everything around it.

Team

Someone who has hired the team you're about to build

22 engineers hired and onboarded, a remote-first migration that measurably raised productivity and satisfaction, and alumni coached into leadership roles abroad. Setting engineering culture is a thing he has done deliberately, not a line he is agreeing to.

Compliance

In-house compliance fluency worth real money

GDPR, HIPAA, WCAG 2 and EU VAT taken through five client organisations firsthand. The kind of knowledge companies usually buy as formal training or discover during an expensive retrofit.

Customers

An engineer you can safely put in front of a customer

Nine years of client-facing delivery across seven countries and a CSPO certification. Extracting build signal from a customer meeting is an existing reflex — you do not need a translation layer between the room and the repo.

Consistency

Consistency demonstrated in public, daily

A 982-day unbroken French streak with A2 grammar explainers published publicly, alongside Medium engineering writing and a Dhaka Tribune policy op-ed. Evidence of the boring virtue that actually ships software over years.

Technical inventory

The stack, row by row

The stack row by row, plus the AI and security receipts underneath the nice-to-haves.

Solve's Must-Haves
  • Python
  • FastAPI
  • PostgreSQL
  • TypeScript
  • React
  • NestJS
  • Full-stack frameworks
AI Systems
  • Claude API / Code / Skills
  • MCP servers
  • Multi-agent orchestration
  • RAG
  • pgvector
  • Evaluation harnesses
Security & Confidentiality
  • OAuth2 / SSO
  • Multi-tenant RBAC
  • Multi-tenant isolation
  • HashiCorp Vault
  • AWS IAM & WAF
  • GDPR
  • HIPAA
AWS DevOps
  • Lambda (container images)
  • ECS
  • EventBridge
  • SQS / SNS
  • CloudFormation (SAML)
  • GitHub Actions
  • Docker
Declaration

London is a deliberate choice, not a tolerated condition.

London is a deliberate choice, not a tolerated condition — Europe's centre of gravity for AI, where the density of serious teams and the competition between them is exactly the pressure he wants on his work. The move is settled ground rather than stated intent: his sister already lives in the city. Remote-first since 2020 across six time zones, so full UK-hours overlap holds from week one, whatever the relocation timeline. More on the city at /geographies/london.

Book 30 minutesEmail
Fauzul Kabir Chowdhury
Founding Full-Stack Engineer (Back-End Leaning)
Solve Intelligence · London, England — on-site; relocating from Dhaka