Founding Engineer (Full-Stack / AI)

SobelAI · Remote — Singapore · Full-time

What is SobelAI

SobelAI helps individual property agents stop losing deals to silence. Agents juggle hundreds of leads across WhatsApp, and the good ones go cold simply because nobody followed up at the right moment with the right message. We built an AI system that re-engages quiet leads automatically — generating personalized, on-brand follow-ups, qualifying ad leads, and passively building a profile of what each lead actually wants — while the agent stays in control of every real conversation.

We're early, the product works, and real agents are using it daily. You'd be employee-level engineer #1 — the person who turns a working product into a category-defining one. This is not a "pick up tickets" role. You'll own systems end to end, set technical direction alongside the founders, and have your fingerprints on every part of the stack.

What you'll actually work on

• The AI message pipeline. A multi-stage system (classify → select strategy → write → guardrail) that decides when to follow up, what to say, and how to say it in each agent's voice. It blends deterministic rules with LLM stages, handles warm/cold/new lead states, cadence scheduling, and per-agent style profiles. There's a constant, fascinating tension here between determinism (reliability, cost, latency) and LLM judgment (nuance, personalization) — you'll be the one drawing those lines.
• Lead understanding. An inbound interpreter that extracts structured facts (budget, areas, intent, eligibility) from messy WhatsApp chatter without hallucinating, and a qualification loop that drives ad leads toward qualified/disqualified/handed-off.
• Messaging infrastructure. Multi-provider WhatsApp integration (Unipile + Meta Cloud API coexistence), lead-source ingestion (Meta lead-gen, 99.co, PropertyGuru), webhooks, and the anti-spam-classifier scheduling that keeps agents' numbers healthy.
• Eval and reliability. As we scale, "did the AI write a good message?" becomes a measurement problem. Building the eval harness, prompt-regression testing, and observability for an LLM-heavy product is wide-open green field — and arguably the highest-leverage work in the company.
• The product surface. The React/Supabase app agents live in every day: inbox, follow-up queue, broadcasts, lead CRM, settings.

What we're looking for

• 3+ years building and shipping production software, with real ownership of features end-to-end (you've designed schemas, written the backend, and built the UI).
• Genuine full-stack range — comfortable in TypeScript/React on the frontend and in backend/data/serverless logic. You don't need to be expert in our exact stack, but you should be the kind of engineer who learns a new layer in days, not months.
• Hands-on LLM/prompt-engineering experience in production (prompt design, structured extraction, eval, guardrails, cost/latency tuning).
• Strong product instincts. You ask "what does the agent actually need?" before "what's the cleanest abstraction?" You're happy talking to users and turning fuzzy feedback into shipped changes.
• A bias for ownership and shipping. In a small team, you'll frequently be the only person who understands a system. You write code that the next reader (human or AI) can navigate, you debug from logs and first principles, and you ship without waiting to be told the next step.
• Comfort with ambiguity and AI's rough edges. LLMs are non-deterministic, sometimes wrong, and improving monthly. You're excited by — not frustrated by — the work of making them reliable in production.
• Strong async communication and timezone overlap with SEA. Remote works because you write clearly, keep work visible, and don't need to be managed.

Bonus points

• Experience with Supabase / Postgres / edge functions, or messaging integrations (WhatsApp, webhooks, async pipelines).
• You've been an early engineer at a startup before and know the difference between "scrappy" and "sloppy."
• Curiosity about the Singapore / SEA property market — it's full of delightful, specific rules our system has to encode.

Our stack

React 18 + TypeScript + Vite · Tailwind + shadcn/Radix · React Query · Supabase (Postgres, RLS, Edge Functions) · LLM pipelines (OpenAI/Claude) · pnpm monorepo. We value the judgment behind the stack over the stack itself — bring strong opinions, hold them loosely.

How we work

Fully remote, small, fast, low-ceremony. Direct access to the founders and to real customers. We prefer application-layer clarity over hidden magic, root-cause fixes over patches, and shipping over perfecting. You'll have unusual autonomy and unusual responsibility.

What you get

• Meaningful founding-engineer equity — you'll own a real slice of what you help build.
• Competitive salary for the region
• A real seat at the table on product and technical direction, at the moment your work compounds the most.