Software Engineer

Software Engineer — Full-Stack, AI-nativeFull-time · Hybrid (Yerevan) · Essential · essentialsln.com Essential curates small teams of world-class engineers to accelerate our clients' innovation cycles. For this role you will join a production AI companion: users talk to an AI avatar via video, voice, or text, powered by a self-hosted LLM stack (vLLM, guard model, RAG), Supabase, Stripe billing, a React/TypeScript frontend, and an iOS app with OTA updates. Stack: TypeScript · React 18 + Vite · Supabase / Deno edge functions · PostgreSQL · self-hosted LLMs (vLLM) · Stripe · LiveKit / WebRTC · Capacitor iOS · AI coding agents. The one non-negotiable: you build with AI agents AI agents are a first-class part of our workflow — Cursor/Claude agents with access to Linear, Supabase, GitHub, logs, and a browser handle implementation, review, and verification, guided by rules and skills the team maintains. We are not looking for "uses autocomplete". We are looking for someone who: • Delegates whole units of work to agents — a ticket, a refactor, an investigation — and scopes, briefs, and constrains them so the output is right the first time more often than not. • Reviews agent output like a senior reviewer — catches hallucinated APIs, silent scope creep, dropped edge cases, and knows when the agent is confidently wrong. • Builds the harness, not just the prompt — maintains agent rules and skills, MCP tooling, and deterministic checks that make agent output verifiable. • Runs agents in parallel and asynchronously, orchestrated rather than babysat. • Knows the limits — why an LLM must never be the sole merge gate, and where humans stay the decision-makers on safety-critical paths. We assess this hands-on: a live delegated task using your own tooling, a review of an agent-produced PR with planted flaws, and a conversation about the harnesses you have built. Your mission Own tickets end-to-end across the frontend, edge functions, database, and LLM pipeline — using AI agents as your primary implementation force multiplier — and raise the quality and speed of the whole team's agent workflow. What you will do Product & platform delivery (~60%) • Deliver tickets across the stack: React/TypeScript UI (chat, voice/video session chrome, onboarding, billing, admin), Supabase edge functions (Deno), Postgres migrations and RLS policies. • Work on the LLM request path: streaming SSE handlers, prompt/context assembly, RAG retrieval, guard-model integration, provider routing and latency work. • Integrate and maintain third-party services for video avatars, voice, Stripe (subscriptions, token top-ups, webhooks), and iOS OTA delivery. • Own token/entitlement correctness: free-trial logic, token grants and debits — code paths where silent bugs cost money or lock users out. Agent-native engineering (~25%) • Ship most work through agents: brief, run, review, iterate. Maintain and extend the agent rules and skills so output converges on team conventions. • Build agent-facing tooling: MCP servers, verification scripts, fixtures and mocks that let agents self-check before a human looks. • Help design the AI feature-test agent pipeline (ticket → diff → test plan → browser run → report) together with our QA engineer. Reliability & operations (~15%) • Keep CI, deploy workflows, and regression runs green; investigate failures using logs, traces, and DB queries. • Participate in LLM ops: vLLM hosts, guard model, model/prompt changes and their quality evals. • Follow and improve the security posture of a safety-critical product. Must-have • 4+ years of professional software engineering, with strong TypeScript on both frontend (React) and backend (Node/Deno or equivalent). • Demonstrable, day-to-day use of AI coding agents (Cursor agents, Claude Code, Codex, or similar) for complete tasks — you will be asked to show this live. • Solid PostgreSQL: schema design, migrations, RLS or equivalent authorization models. • Experience with streaming / real-time systems (SSE, WebSockets, WebRTC, or audio/video pipelines) — buffering, backpressure, partial-chunk handling. • Production LLM integration experience (prompting, structured outputs, streaming, evaluation) and a grounded view of non-determinism. • Comfortable with Git-based ticket flows, code review, CI/CD (GitHub Actions), and writing tests where they matter. • Fluent written English; German is a plus (the product's primary locale is DE). Nice-to-have • Supabase (edge functions, pg_net, pg_cron, auth JWT internals) · Stripe billing · Capacitor / iOS OTA pipelines. • Self-hosting LLMs (vLLM, GPU sizing, guard models), RAG/embeddings, Python for eval scripts · Playwright. • Prior work in health, mental health, or other regulated / safety-critical domains. Success after 90 days • Multiple tickets shipped to production independently, most implemented predominantly through agents you directed. • At least one measurable improvement to the agent harness (rules, skills, MCP tooling, or fixtures) adopted by the team. • One production investigation handled end-to-end: logs → root cause → fix → regression coverage. What we offer • Hybrid, async-friendly, small team with direct product impact. • A mature agent-native engineering setup you can shape, not fight: MCP servers, skills, CI, and regression automation already in place. • Real ownership — your tickets ship to production independently, usually within days. • Budget for AI tooling (Cursor, Claude, model APIs) is part of the job, not a perk. How to apply Send your CV plus a short note on the most impressive thing you have shipped with AI agents to jobs@essentialsln.com. Full description: https://www.essentialsln.com/careers/software-engineer