Full stack engineering for products that have to hold up.
Five years of it in a bank, where the frontend was eight micro-apps and the backend was a hundred endpoints nobody was allowed to get wrong. I take contract work across that whole span - the architecture, the database behind it, and the LLM features that have to survive production.
Balance
Shipping production software
Independently deployable
Across two design systems
Secured with RBAC
Delivered on the LMS platform
Services
Four disciplines, taken together or one at a time. Each is something I have already shipped and can point at - nothing here is offered on the strength of having read about it.
Frontend Architecture
One repository, several teams, and every release waiting on the slowest of them.
Drawing the boundaries, standing up the runtime that resolves them, and migrating the first domains across without a rewrite freeze. Single-SPA or Module Federation on evidence, not preference - I have shipped both and they fail differently.
- Boundary map and migration sequence
- Host/shell and runtime resolution
- First two domains migrated end to end
- Design system the boundaries share
- Written decision records
Backend & Data
An interface that has outgrown the thing behind it - slow pages, numbers that disagree, and a permission check that only exists in the UI.
Schema first, then the queries the product actually needs. Aggregation belongs in SQL rather than in the browser, authorization is re-verified on the server inside every mutation rather than hidden in the interface, and money stays in integer cents until the moment it is printed.
- Schema, migrations and seed data
- Server-side data layer, typed end to end
- Auth, sessions and role-based access
- Validation at every mutation boundary
- Deploy and connection pooling you own
AI Features, Production-grade
A demo that impressed the room and has been stuck there for two quarters.
Retrieval that returns the right document, agents that stop when they should, and an eval loop that tells you whether last week's prompt change helped. Versioned prompts, logged cost, measured output.
- Retrieval pipeline with a measured baseline
- Tool routing and human-in-the-loop gates
- Streaming UI
- Prompt version registry
- LLM-as-judge eval harness
Delivery & Operations
Code that works on a laptop and nobody is confident enough to release on a Friday.
The unglamorous half of shipping into a bank: a pipeline that gates quality before production rather than after it, and enough observability to know when it did not. Usually taken alongside one of the three above rather than on its own.
- CI that gates on types, lint and tests
- Containerised build and serve
- Static analysis and quality gates
- Error monitoring wired to real releases
- Runbook for the team that inherits it
Architecture Review
A decision that is expensive to reverse, and nobody outside the team to argue it with. I read the codebase, interview the people carrying it, and write up what I would change, in what order, and what I would leave alone. A document you own and can act on without me.
- Codebase and dependency read
- Interviews with the team
- Findings ranked by cost of delay
- Sequenced recommendation
- One call to argue it through
Selected work
The engagements behind the services above. The enterprise work is under NDA, so what is shown is the architecture and the measured outcome - drawn from scratch, no client assets, and yours would be treated the same way. Meridian is mine and carries no such restriction: it is running, and you can sign in and read the queries yourself.
How it runs
The same four steps whichever service it is. Nothing starts before the scope is written down.
- 01
Call
Thirty minutes. What you are building, what is blocking it, and whether I am the wrong person - I will say so if I am.
- 02
Written scope
Milestones, what is included, and what is explicitly not. Nothing starts on a verbal agreement.
- 03
Build
Your repository from the first commit, reviewed by your team, demoed weekly. No black box, no big reveal.
- 04
Handover
Decision records for everything non-obvious, and a team that can carry it without me. Availability for questions after, at no charge.
Audit trail
Frontend Developer
Banking ERP, an enterprise LMS, and an internal portal migration for a Korean broadcaster.
Web Developer
Crypto token swap interfaces and Web3 wallet integration.
Full Stack Developer Program
Intensive bootcamp - the pivot from mathematics into software.
B.Sc. Mathematics
Linear algebra and probability - the same machinery underneath embeddings, vector search and model evaluation.
Contact
Building something that needs to hold up?
Tell me the problem and the deadline. A reply inside two working days, and a straight answer about fit before anything else.