Candidate Requirements

What we're looking for

 **A computer scientist by training or by demonstrable depth.** If you have a CS degree, great. If you don't, you should be able to hold a 30-minute conversation about a non-trivial algorithm or system-design problem without assistance. Self-taught is fine.

 **Strong production Python.** Django, async, the standard library beyond `os` and `json`. You can read a stack trace, profile a slow path, and write tests that catch regressions.

 **Database fluency.** PostgreSQL specifically — schema design, query optimisation, hand-written migrations, transactional discipline. You can read a slow `EXPLAIN ANALYZE`. You've worked with a vector database (pgvector or equivalent).

 **Cloud + container** AWS (S3, RDS, Cognito, SES), Cloudflare and Docker / Docker Compose.

 **CI/CD with GitHub Actions.** You've built and maintained workflows that test, build, and deploy production services.

 **System architecture instinct.** When you see a multi-component pipeline, you immediately reason about failure boundaries, idempotency, where backpressure should live, what blocks the user-visible loop. You understand patterns like microservice architecture, Mediator patterns, CQRS, event-driven inter-service communication, and you know when each is and isn't the right tool.

 **Pragmatic frontend.** You can read React + TypeScript, debug a controlled-form state issue, wire an SSE consumer to a UI, ship a small UX fix. You won't be the design lead. However, you will be expected to stage the frontend pieces of your own backend features to the front-end team.

 **Tooling judgement.** You use Claude Code / Cursor / Copilot as a force multiplier, not as a brain. You can tell when an AI suggestion is wrong (think first, the have the LLM help, not the other way around). You manage your own dev CI.

 **South African work-time overlap.** We run on SA business hours.

 Strong nice-to-haves

- Production experience with vector search, hybrid retrieval (vector + keyword + reranker), or RAG pipelines.

- Domain-Driven Design experience with bounded contexts and a real domain event bus.

- Streaming (SSE / WebSockets) shipped to a real frontend, including the proxy-timeout / heartbeat playbook.

- Microkernel / plug-in patterns where business logic is externalised as configuration (YAML, JSON-Schema, similar).

- POPIA / GDPR / PCI-DSS compliance work — data subject rights, retention, envelope encryption with cloud KMS.

- Document-processing pipelines (OCR, PDF extraction, DOCX generation, in-browser editors).

- Legal-tech, regtech, fintech, healthtech, or any other regulated-domain background.

 How we work

 - **Pragmatism over dogma.** We default to the simplest thing that works now but also in the future. We don't add abstractions speculatively.

- **Telemetry-driven debugging.** When something's wrong in production, the answer should be in the dashboard, not on someone's laptop.

- **The lead owns the legal accuracy.** You don't write South African legal content from general knowledge; you ask. Legal accuracy is a professional-misconduct-level constraint, not a nice-to-have.

- **Direct communication.** Short, factual, friendly. You raise blockers early. You stage before perfect, then iterate.

- **Real startup hours.** We try to keep evenings and weekends free, and we usually succeed. There will be weeks where a launch, a deadline, or a production issue means longer hours. We are honest about that up front.

 Don't apply if

 - You let Claude / Cursor / Copilot **think** for you. Use them, don't outsource your judgement to them.

- You haven't managed your own dev CI.

- You think "DevOps is someone else's problem." It isn't here — you'll touch Docker, deploy scripts, observability config.

- You don't read code carefully.

 Process

 1. CV + a short note on **the most interesting non-trivial system you've built** (one paragraph; tell us what was hard about it, not what it did).

2. Technical screen — we read code together; you walk us through reasoning; show us how you use AI as a tool and not a replacement.

3. Final round — system design conversation, plus team fit.

4. Offer.

 We aim to close in two weeks from first contact for candidates we're serious about.