01 / Keanu
Keanu Chetty
I work across IT systems, software, QA, integrations and implementation — figuring out what people need, translating it into something technical, and helping carry it through build, test, rollout and support.
Status — Open to opportunitiesCharleston, SC / Remote / New Jersey
02 / Signal
Numbers from systems I actually administered, tested or launched — not a skills bar chart.
Tenant administration — identity, mailboxes, licensing
Sites supported at the network's peak
Locations taken from onboarding to live
Defects identified, documented and driven to resolution
Onboarding time after the documentation rebuild
03 / Professional systems
Day-to-day this is systems administration and technical operations — the infrastructure, support and QA that keeps a franchise network running. It is also why the operational half of a build does not get forgotten.
03.1
Microsoft 365 tenant administration across the franchise network — identity, mailboxes and licensing. Migrated team operations onto Teams and built a SharePoint intranet so documentation had one home.
Microsoft 365 · Entra IDExchange · TeamsSharePoint
03.2
Frontline application, email and system-configuration support for a network that scaled to 50+ locations at its peak, plus the onboarding work that took 12 of those locations live.
Tier 1–2 supportEmail & DNS troubleshootingLocation onboarding
03.3
Pre-launch QA for a multi-tenant SaaS platform: seven test tenants provisioned by hand to stress the system end to end, confirming customer, order and franchise data never crossed between accounts and that every transaction charged the right account.
Multi-tenant isolation testingStripe payment routingIntegration & checkout QA
03.4
Authored end-to-end support documentation and rebuilt onboarding around it, cutting training from three hours to ninety minutes. Administered Linux systems and automated routine maintenance with Bash.
Scribe documentationOnboarding & trainingLinux · Bash automation
03.5
Testing and supporting the seams between the platform and the systems around it — distributor and aggregator catalogues, field-service scheduling, fleet telematics, payments and messaging. Most of the work lives at the boundary, where a failure belongs to neither system on its own.
Housecall Pro · SamsaraTireHub · ATDStripe Connect · Twilio · SendGridREST APIs · Webhooks
04 / Selected work
Four properties I designed, built and still operate — domains, hosting, deliverability and the fixes nobody budgeted for. Open a row for the case and a live embed of the site itself.
05 / Current builds
Most of my personal projects start because I run into a problem I want to solve. Some become finished tools. Some turn into experiments. Some teach me enough that I don't need to finish them.
Build.01ACTIVE
Four surfaces on one codebase.
A customer-facing storefront, an operator back-office, a field mobile app and a platform admin layer, all reading one data model. The hard part isn't the interface — it's tenancy: making sure one tenant's data cannot leak into another's under any query path, while three commerce systems each want to own the customer record.
Next.js · Supabase (RLS) · Medusa.js · Stripe Connect · Twilio · SendGrid · Vercel
Build.02ARCHIVED
Paused at phase ten, with the approval loop working end to end on paper trading.
A personal stock tracker on the Alpaca API: portfolio tracking, event alerts and AI-assisted research. Deliberately not autonomous — every order needed explicit human approval, routed through GitHub issues so it worked from a phone and left an audit trail. Real money sat behind safety rails and a fourteen-day clean paper run.
Python · Alpaca · Supabase · Anthropic · Finnhub
Build.03ACTIVE
Twelve specialists, one orchestrator, and constraints the agents cannot argue with.
A team of specialised coding agents — architecture, security, database, review, QA, docs — each with its own tool whitelist and model tier, and an orchestrator that is the only thing allowed to start them. The reviewers hold no write or edit tools, so the review step cannot quietly become an editing step. The rule I am proudest of is duller: acceptance tests are human-authored and committed before any agent writes the implementation, and a change that supplies both is rejected.
Claude Code subagents · Model tiering · Spec-driven · Human approval gates
Build.04ACTIVE
Five hundred applications a month is not a number you review by hand.
A national-retailer partnership means the network has to find, vet and approve far more independent installers than anyone can process by hand — the forecast is around five hundred applications a month. Two halves: the approval process that decides whether an applicant qualifies, and the onboarding wizard that carries the ones who do into the field-service platform. Intake, identity verification, background and motor-vehicle checks, insurance validation and e-signature each become an approval state rather than an email thread, with the downstream onboarding handed off over an API once somebody clears. The decisions that matter are not on the happy path: which vendor owns which check, who owns the data afterwards, and how much sensitive personal information the system should still be holding once a decision has been made. I am on the team building both halves.
Approval-state workflow · Identity & background verification · E-signature · API handoff
Build.05PROTOTYPE
A watchlist is easy. Two people agreeing on one film is the actual problem.
Tracks what you have watched and suggests what to watch next, which is the easy half. The half worth building was group choice: each person accumulates their own profile, and a recommendation for two people is the overlap between their profiles rather than the average of them — averaging two tastes reliably produces a film neither person actually wants. Tracking and single-viewer recommendations ran; the multi-person intersection was the part still being built when it stopped.
Build.06PROTOTYPE
Recipes into a week, a week into one shopping list, without retyping anything.
The recipe form is not the interesting part. What happens between the plan and the list is: two recipes wanting onions on different nights should arrive as one line, a recipe scaled from four servings to six has to carry its quantities up with it, and units have to survive being added together at all. The whole loop ran end to end — plan a week, get a grouped list — it just never got past rough.
06 / The system
Nothing on this page is a one-off.
Every site is built from the same Next.js template and runs on shared infrastructure — one database behind all of them with row-level tenancy, one form pipeline, one deploy path where a push to main is a production release. Schema changes are written down as architecture decisions before they happen.
The portal is the front door: clients sign in with a magic link and see their site and nothing else — leads, change requests, publishing, analytics. Whatever ships next lands behind the same login.
One template · Shared Supabase (RLS) · SendGrid · GA4 · Vercel · ADR-governed schema
07 / How I work
I take projects from design to production and stay responsible for them afterward. IT systems administration is the day job — Microsoft 365, infrastructure, support — which is why the operational half of a website doesn't get forgotten here.
01
Sit with the people who use the thing and work out what they actually need, which is rarely what was asked for.
02
Design and front-end, application logic, database schema and access control, third-party integrations.
03
Provision real tenants, walk the flows, confirm data and payments stay where they belong.
04
Domains and DNS, hosting, email authentication, analytics, search.
05
Write the documentation and walk people through it, so it doesn't come back as a ticket.
06
Monitoring, updates, backups, and the small fixes nobody budgeted for.
07
Watch what breaks or gets asked twice, then go fix the cause rather than the symptom.
08 / Before this
Operations, a development internship, a computer science degree and a year teaching kids to code. Different rooms, same habit: work out how the thing actually runs, then point technology at it.
2021
Operations · Automation
Operations Lead and IT support. Python turned up before the job title did — scripting against inventory, alongside scheduling and frontline support.
Jacksonville, FL
2022
Software · Process
A summer development internship: Agile and Scrum, code review, and the software development life cycle seen from inside a working team.
Portland, OR
2022
Systems · Security
Computer Science with a cybersecurity minor. SQL databases, C# web applications, project planning, and the first intrusion-detection labs.
Jacksonville, FL
2023
Teaching · Communication
Lead Sensei. Taught JavaScript, robotics and cryptography to students of wildly different ages — where explaining technical work to non-technical people started.
Jacksonville, FL · Livingston, NJ
09 / Field notes
A public notebook for commands, fixes, experiments, certification notes, API discoveries and things I know I'll want to remember later.
Currently learningSC-900Microsoft SecurityCybersecurity
N-008FIX
I had been describing my review agents as read-only because their tool lists omit write and edit. Re-reading the files, both still carry an unrestricted shell — and a shell is a write primitive. The guarantee I thought was structural was really a convention with a gap in it, and the one agent I had genuinely locked to read, search and glob was the one I never installed.
agentssecurityaireview
N-010LEARN
An eighty-thousand-token specification loaded wholesale into every session crowds out the task itself. Replacing it with a short resident index of line anchors, plus explicit cues telling the agent when it must go fetch, cut the resident load sharply. The catch is that a stale anchor is worse than a missing one — no fetch leaves the agent knowingly uninformed, a wrong fetch hands it plausible material and no reason to doubt it.
agentscontextdocumentationai
N-009TEST
Acceptance tests are written by a human and committed before any agent starts implementing, and a change that supplies both the implementation and its only acceptance tests is rejected. An agent that writes its own definition of done will meet it. The rule is dull, slows things down, and is the one I would keep if I could keep only one.
testingaiprocesssafety
N-007THOUGHT
Architecture, security review and code review run on the strongest model; documentation runs on the cheapest; implementation sits in the middle. It is the same judgement as deciding which changes need a second reviewer — cost tracks the blast radius of a bad answer, not the volume of output.
agentsaicostreview
10 / Lab
Security work I do on my own time, in sandboxes built for it. It isn't the job title — it's why I think about failure modes and access paths when I build something.
Lab.01
Worked through multiple HTB machines — reconnaissance, exploitation and privilege escalation — in a legal, sandboxed environment built for exactly that.
Recon · Exploitation · Privilege escalation
Lab.02
Tested the Hak5 OMG Cable and a Flipper Zero in a personal lab to understand how physical-access attacks actually work — and what defends against them.
Keystroke injection · RFID · Sub-GHz
Lab.03
Coursework toward a SOC analyst certification: security monitoring, log analysis and the incident-response fundamentals that sit under all of it.
Monitoring · Log analysis · Incident response
Lab.04
Coursework toward SC-900 — Microsoft security, compliance and identity fundamentals. The formal counterpart to the Microsoft 365 tenant administration that is already the day job.
Security · Compliance · Identity
11 / Outside work
I grew up in Johannesburg and moved to the U.S. at nineteen for university. I was the unofficial tech support at home long before it was a job title — the person who got called when something stopped working and nobody knew why.
Later I built my own gaming PC, which is where a lot of this started: wanting to know how the pieces fit together badly enough to take them apart. That has not really changed. It is the same instinct that makes an unfamiliar system interesting rather than intimidating.
Outside of work, I'm usually messing with my car, playing with Kona, gaming, printing something I probably don't need, or researching a new gadget.
Currently interested in
12 / Contact
Hiring, contracting, or comparing notes on something technical — I reply within one business day.
keanu@zerange.com
Charleston, SC / Remote / New Jersey