The full-stack engineer who has already built both halves.
Video at scale, and AI-native features on top of it. Shipped separately, in production, before this application existed.
Integrated evidence to prevent, detect, and respond to every objection
Mapped line by line against Coram's own bullets — the video-at-scale backend, the AI-native feature work, the frontend stack, and the cloud-native infra bar. Honest verdicts, including the two that aren't strong.
type to filter the feed · press Enter to ask Fauzul
Build and maintain backend services that ingest, process, and serve video and sensor data at scale
is the direct precedent: a cross-platform video social platform serving 10,000+ active users, where Fauzul built the event-driven media pipeline — AWS SQS queues, SNS notifications and dynamically-scaled EC2 encoder instances transcoding user video uploads, with automatic audio-video sync-offset correction on ingest. runs the same shape today under Go orchestration: FFmpeg concatenation of 16-bit PCM @ 24kHz into long-form episodes, transcoded to WAV/MP3/OGG and served from Cloudflare R2 behind self-healing RSS feeds. The cost dimension is lived too — at he split media workloads on purpose, running FFmpeg conversion out of Mumbai for compute cost while keeping client-serving endpoints in California for latency.
Ship AI-native features end to end — natural-language search, real-time detections, and agentic workflows built on multimodal LLMs and computer vision
The natural-language and agentic two-thirds are daily practice. VisaPros runs six parallel FastAPI agents on Gemini with live Google Search grounding, synthesising a ranked roadmap rather than a single completion. Global Jute Trading runs a production CMS with Claude as the harness — plain-English requests become reviewed, CI/CD-deployed pull requests on Cloudflare Workers. Retrieval and ranking are shipped surfaces (pgvector, MeiliSearch), and agent evaluation harnesses — accuracy, safety, latency — are part of how he ships, not an afterthought.
Computer vision is the honest third. The closest owned work is ML inference inside a media pipeline — 's AI vocal/instrument separation isolating stems from licensed masters — plus generative-vision prototyping at . Consuming a detection model's output, versioning it, and building the product surface over it is familiar; training or tuning the CV models themselves is not his lane, and on this team it shouldn't need to be.
Implement user-facing frontends in TypeScript/React, working closely with product
Coram's exact frontend line, row for row: TypeScript, React 19, TanStack Query, Tailwind v4 — the stack behind 's admin panel and B2B checkout engine (15,000+ active users, GDPR and EU VAT compliance built into the payment flows) and 's marketplace surfaces. Design-to-code is one hand, not a handoff: Figma through Storybook through production, with WCAG 2 accessibility shipped across five client organisations.
Strong experience with Golang, Python, or C++
Go carries 's central orchestration and 's ordering service on on-demand Cloud Run, with ConnectRPC for typed service boundaries. Python carries ' six FastAPI agent services, backed by a Kaggle Python for Data Science certification. C++ is not part of his stack — the JD reads these as alternatives, and Go plus Python is the pair he brings.
Experience designing and building scalable backend systems and APIs
Anygum is the load-bearing example — a multi-tenant software marketplace he conceived and architected, where providers register once and consumers discover under federated RBAC and OIDC multi-tenancy (Auth0) with isolated per-org instances. , the ERP designed end to end for multi-organisation business operations, runs on that backbone across client organisations. Underneath: NestJS, PostgreSQL, Redis, REST and ConnectRPC, and event-driven AWS (Lambda, SQS/SNS, EventBridge).
Design scalable APIs and systems that interface with ML pipelines and edge device fleets
The ML-pipeline side is real: 's separation pipeline and 's multi-model orchestration both sit behind APIs he designed, with queue-backed backpressure rather than synchronous calls into inference. Edge compute is shipped daily on Cloudflare Workers (, fauzul.com).
A fleet of physical cameras is genuinely new — his edge experience is compute-at-the-edge, not device management, provisioning, or OTA. The transferable half is the queue-and-reconcile discipline that already survives unreliable producers; the camera-fleet specifics are a first-months ramp he'd rather name than paper over.
Tackle problems involving distributed systems, real-time processing, and high availability
AudioBin is a real-time collaborative browser DAW with custom WebSocket synchronisation holding sub-50ms collaborative state — a four-year engagement with , SkyTracks & SoftCollab in Laval and Montreal, engineered alongside the client's technical CTO. 's real-time clinical dashboards run live today at 64 doctors, 3,242 patients, 13,340 schedules across 7 centers. Event-driven decoupling (SQS/SNS/EventBridge, n8n DAG orchestration) is the standing pattern, not a one-off.
Comfort with distributed systems, containers, and cloud-native infrastructure (Kubernetes on AWS, Pulumi for IaC)
Containers and AWS are deep and hands-on: Docker daily, and production delivery across EC2, ECS, Lambda, EventBridge, DynamoDB, SQS/SNS, CloudFormation and IAM/WAF, plus GCP Cloud Run. Infrastructure-as-code is a genuine habit rather than a checkbox — he ran CloudFormation SAML-based provisioning in production and hosted a public GitOps-with-Terraform webinar series.
Kubernetes sits at familiarity, not production ownership, and Pulumi he has not shipped. Both are thin layers over instincts already in daily use — declarative infra, GitOps, containerised services — so the ramp is real but short. Call it three to four weeks to be useful on Coram's cluster and trusted in its IaC reviews.
Fluency with AI-assisted development and opinions on where it works well
This is the strongest single line on the page. Anthropic-certified — Claude Code 101 and Introduction to Agent Skills — and actively working through further Anthropic tracks. The opinions are published and load-bearing: WSL Linker was root-caused and shipped as a scoped MIT tool the same week an IDE update broke a team's workflow; TagRamp's AI layer was rearchitected mid-flight from client-side to backend-driven the moment enterprise security requirements made the original model unworkable. He writes about it publicly too, including a Medium piece on developer productivity in the AI era. The compounding case is Wiregent: 195 commits in eight days, agent-led, with the test count growing faster than the source count — velocity that holds its shape because the guardrails were written first.
Previous startup experience building systems from scratch, or scale-up experience knowing what good looks like
Co-founded ELO in 2017 and has been there 9 yrs 1 mo — UX Engineer to Principal Staff SWE across the IC arc, then Engineering Manager, now CIO. Grew the team to 22 people and then ran it distributed remote-first from 2020 (+50% productivity, +100% satisfaction), replaced manual deploys with GitHub Actions and AWS pipelines for a 75% reduction in time-to-market, and delivered 0→1 across five verticals — music-tech, gaming and esports, hospitality, health-tech and professional networks. Scale-up receipts include , an client for 6+ years through to its acquisition by Loopcloud in 2026. The current-velocity proof is Wiregent, a stealth YC project (Fall 2026 batch): in eight consecutive days it went from 101 to 222 source files and 17 to 47 test files across 195 commits — an agent-negotiation platform whose settlement clears only through a double-entry, hash-chained, append-only ledger.
A strong understanding of system performance, reliability, and clean architecture
Performance shows up as specific calls with reasons attached: EC2 fleets justified with instance comparisons and split across regions on purpose, Lambda chosen where cold-start reliability beats the alternatives, and an early adoption of container images in Lambda that still runs for environment consistency and cost. Clean architecture is a published position — his Medium piece on monorepos, multi-project components and styling consistency is the monorepo ( and ) written up. Reliability was made structural: GDPR, HIPAA and WCAG 2 compliance led into products across 5 client organisations. Wiregent's ledger is the sharpest recent example of the same instinct: one sole-writer module, balanced legs or the whole transaction is rejected, integer minor units rather than float money, balances derived and never stored, and no UPDATE/DELETE path at all — corrections are compensating entries.
A Computer Science degree or similar from a top university
Field of Study: Computer Science & Engineering · North South University, Dhaka — left before completing to build, and has been shipping since. This is the one line on the page he cannot answer with a yes.
There is no ramp for a credential — only the substitute record. Ten years of production systems, a co-founded company grown to 22, a video platform at 10,000+ users, a HIPAA platform live across 7 centers, and an acquired client product. If a completed degree from a top university is a hard filter, this is the right place to stop; if 'or similar' can mean the work, the work is all documented and linked.
Four things that are true before the interview starts
Every claim below is checkable — against a running business, a live platform, or a shipped repo.
Video at scale is prior work, not a first attempt
— a video social platform for 10,000+ active users — ran on FFmpeg, SQS/SNS queues and dynamically-scaled EC2 encoders, with automatic audio-video sync-offset correction on ingest and sub-50ms playback sync on the client. runs Go orchestration over FFmpeg pipelines to Cloudflare R2 daily. At , media conversion was deliberately split by region — Mumbai for compute cost, California for serving latency.
AI-native features shipped end to end, in production
's CMS uses Claude as the harness — plain-English requests become reviewed, CI/CD-deployed pull requests on Cloudflare Workers, in production for a 26-year-old export house. runs six parallel FastAPI agents with live search grounding. Anthropic-certified (Claude Code 101, Introduction to Agent Skills) plus the Google × Kaggle 5-Day AI Agents Intensive.
One hand from Figma to the API to the deploy
TypeScript, React 19, TanStack Query and Tailwind v4 on the front; NestJS, Go, FastAPI, PostgreSQL and Redis behind. is his own architecture — federated RBAC and OIDC multi-tenancy with isolated per-org instances, with running on it. 's admin panel and B2B checkout served 15,000+ users with GDPR and EU VAT compliance built into the payment flows.
Startup-from-scratch, then the discipline that keeps it alive
Co-founded in 2017, grew it to 22 people, moved it remote-first in 2020, and cut time-to-market 75% with GitHub Actions and AWS pipelines. Led products into GDPR, HIPAA and WCAG 2 compliance across 5 client organisations — the reliability habits Coram will want once Fortune 500 deployments outnumber the local ones.
Ships fast without lying about what shipped
(stealth YC project, Fall 2026) went from 101 to 222 source files in eight days — 195 commits, tests growing from 17 to 47 files alongside them. It also ships a simulation ledger: a row-by-row inventory grading every capability as real, simulated, demo-theatre, or absent and reported as absent — on the argument that some things are worth more as a stated gap than as a plausible number. One of its own commits reads “the live site was overselling what it cannot do and hiding what it can.” That is the same doctrine this page runs on.
Compliance shipped as a product surface, not a retrofit
Led products into all three across 5 client organisations — baked into user-facing flows from day one. For a platform heading into Fortune 500 and school-district deployments, that is a procurement cycle that closes rather than one that stalls.
The objections a sharp Coram interviewer would actually raise
The direct objections a sharp Coram engineering interviewer would raise — the degree line, the Kubernetes/Pulumi ramp, and the computer-vision distance — answered straight.
The JD asks for a CS degree from a top university. You don't have one.
Correct — I studied Computer Science & Engineering at North South University in Dhaka and left before completing it to build. I won't argue the line; it's the one requirement on your list I can't answer with a yes. What I'd offer instead is the record it was meant to predict: nine years co-founding and growing it to 22 people, a video platform at 10,000+ users with its own transcoding pipeline, a HIPAA platform live across 7 centers, and a client product that got acquired. If the credential is a hard filter for Coram, I'd genuinely rather know early than spend your team's interview cycles.
You run Kubernetes on AWS with Pulumi. Have you actually operated either?
No — and I'd rather say that than perform confidence. Kubernetes is familiarity, not production ownership; everything I've shipped runs on Docker with AWS (ECS, Lambda, EventBridge, SQS/SNS, CloudFormation) or GCP Cloud Run. Pulumi I haven't shipped at all. What I do bring is the declarative-infra habit underneath both — I ran CloudFormation SAML provisioning in production and hosted a public GitOps-with-Terraform webinar series, so the mental model of infrastructure as reviewed code is already how I work. Realistically that's three to four weeks to be useful on your cluster and trusted in IaC review, not three months.
This is a video security platform. What's your computer vision experience?
Adjacent, not owned — and the distinction matters. I've shipped ML inference inside a media pipeline: 's AI vocal and instrument separation, pulling stems out of licensed masters at scale, and generative-vision prototyping at that replaced $10,000+ of food photography. So I know how to put a model behind an API, version it, handle the failure modes, and build a product surface people trust on top of its output. What I haven't done is train or tune detection models, and on a team with your AI background I'd be the wrong person to try. I'd rather integrate your models well than duplicate them badly.
Your founders came out of Zoox, Lyft and Stanford self-driving. Won't you be out of your depth?
On autonomy research, yes, and I'd want to be — that's the reason to join rather than a reason not to. Where I'm not out of my depth is the part that turns that research into a product people can use every day: the ingest pipeline that doesn't fall over, the API that survives a flaky edge fleet, the frontend that makes a detection legible in two seconds, and the delivery discipline that gets it shipped weekly. I've been the person who takes ambiguous, technically-deep work and puts it in users' hands for a decade, across five verticals. That's the seam I'd be working.
Why leave a CIO role at your own company for an IC engineering seat?
Because I never stopped being the person who writes the code, and the title has started to get in the way. Everything I've shipped in the past two years — , , , , the Claude-powered CMS for — I built with my own hands, in parallel with running . Coram is at the stage I actually enjoy most: Series B, revenue, real customers, and a core engineering team small enough that one person can own a feature from the ingest queue to the pixel. I'd rather do that on a product I believe in than manage a calendar around it.
You're not in the UK. What's the relocation and work-authorisation picture?
London is the move I'm already making, and it isn't abstract. Professionally it's the room I want to be in — Europe's densest concentration of AI labs, applied-AI startups and hackathons is there, and that competitive pressure is the reason I want it, not a cost I'm accepting. It's the environment that pushes my work forward. Personally, my sister lives there, so the move has a landing place as well as a reason. I'd need UK Skilled Worker sponsorship, which I've researched and am prepared for, and I've run remote-first teams across six time zones since 2020 — so I can be fully productive from day one and relocate on whatever timeline the visa and your team need.
3 ramps, with honest timelines
The lit one is the shortest and the most load-bearing for this role.
Kubernetes ownership and Pulumi IaC
Docker is daily and AWS is deep (ECS, Lambda, EventBridge, SQS/SNS, CloudFormation, IAM/WAF), with GCP Cloud Run alongside it — and IaC is already a habit, from production CloudFormation SAML provisioning to a public GitOps-with-Terraform webinar series. Kubernetes is at familiarity rather than production ownership and Pulumi is unshipped, but both are thin layers over declarative-infra instincts already in use daily.
Computer vision and multimodal detection models
He has shipped ML inference inside a media pipeline — 's AI vocal/instrument separation isolating stems from licensed masters — and generative-vision prototyping at that cut $10,000+ in photography costs. Consuming detections, versioning models behind an API, and building the product surface over them is the transferable part; the CV modelling itself belongs to Coram's AI team, and he would rather integrate their work well than duplicate it badly.
Edge camera-fleet management (provisioning, OTA, device health)
Edge compute is daily on Cloudflare Workers, and the queue-and-reconcile discipline for unreliable producers is already proven in 's upload pipeline. Physical device fleets — provisioning, firmware rollout, health telemetry — are new surface area, best learned against Coram's actual fleet rather than claimed in advance.
A completed Computer Science degree from a top university
Not something a teammate can cover and not something time fixes — the degree at North South University was a field of study he left to build. The offsetting record is ten years of production systems, a company co-founded and grown to 22, and shipped platforms at 10,000+ and 15,000+ users. If the credential is a hard filter, that is a fair place for Coram to stop.
The stack, row by row
The stack row by row, with the video-pipeline and agentic receipts sitting under the bullets they actually answer.
What comes with him that the job description didn’t ask for
Habits that show up regardless of the title.
The investigation closes. The conversation is the next step.
London is already the plan — and it is the AI hub Fauzul wants to be building in. Europe's densest concentration of AI labs, applied-AI startups and hackathons sits there, and that competitive pressure is the draw rather than the cost: it is what pushes the work forward. His sister lives in London too, so the move has a landing place attached as well as a reason. UK Skilled Worker sponsorship researched and welcome; remote-first across six time zones since 2020 means full productivity from day one while the visa runs. See the London relocation brief.