Remote Β· Full-Time Β· English async Β· Time Zones: CET +/- 2 hours Β· Reports to Founder
βββββββββββββββββββββββ
ABOUT THE WORK
Riffbird is one of the largest social recruiting agencies in DACH, with 600 clients across Austria, Germany, and Switzerland. We run paid social campaigns, pre-qualify applicants on the phone, and operate the full hiring funnel for our clients.
The service works. The infrastructure does not. Today it runs on a patchwork of external SaaS tools held together by manual operator work. We're now bringing the operational layer in-house, piece by piece, and we're building it so LLM agents can do most of the work that humans currently do by clicking around.
You're coming in to build that layer. It's not a greenfield reinvention of recruiting software. It's the platform that turns an already-working service into something agents can run.
βββββββββββββββββββββββ
WHAT YOU'LL BUILD
One platform, built around a single asset and a set of agents that operate on it.
THE DATA LAYER (the most important thing we own)
A unified candidate and client database. Every campaign, every screening call, every interview outcome feeds it. A candidate who doesn't fit Client A stays discoverable for Client B with proper consent. Full-text plus semantic search. This is the part of the business that's hardest to replicate from outside, and everything else on the platform exists to feed it, query it, or act on it.
THE AGENT WORKFORCE
The operational work we do today, moved one capability at a time onto agents that can run with human supervision:
β’ Ad creative generation (already live with Claude Code, gets expanded and integrated)
β’ Recruiting landing page generation, per client and per role
β’ Candidate research, active sourcing, and headhunting (LinkedIn, niche platforms, our own pool)
β’ Outbound qualification calls (telephony integration, region-by-region pitch and screening logic, full transcript and outcome tracking)
β’ Outreach and follow-up drafting
β’ Pipeline operations: moving candidates through stages, flagging stalled ones
β’ Reporting and anomaly detection across ad spend, funnel, and conversion
Our operators stay in charge of the work, the agents scale what they do. A capability moves from human-driven to agent-driven only when the evals back it up.
THE CLIENT WORKSPACE
Where our 600+ clients see their pipelines, get their reports, and talk to us. Multi-tenant, fast, clean. Today this is split across a stack of external tools and Slack, getting it in-house is step one.
THE AGENT PLATFORM
Typed, scoped tools, audit trails, eval pipelines, the infrastructure that lets us ship new agent capabilities without breaking client work. GDPR-clean throughout (German clients, German applicants, non-negotiable).
βββββββββββββββββββββββ
WHAT YOU BRING
We don't care which language or framework you pick. We care that you've done the hard parts somewhere before:
β’ 5+ years shipping production systems where state, queues, webhooks, and idempotency actually mattered, not just CRUD over a single table
β’ Built and operated multi-tenant SaaS with a real permission model, not just WHERE tenant_id = ? stitched into every query
β’ Shipped LLM features in production where it had to work: RAG that retrieved the right thing, tool-calling that didn't hallucinate the schema, evals that caught regressions before users did
β’ Designed APIs and data models that agents can use without supervision, or at least understand what changes when the consumer is an LLM instead of a frontend
β’ Operate well in EU/GDPR territory, or have the discipline to learn it without us getting fined (in cooperation with our GDPR-expert)
β’ Async-remote operator: you write things down, you disagree in writing, you don't need standups
β’ Comfortable working directly with a founder who builds product himself with AI tools but isn't going to sparring-partner you on systems design. You own the engineering decisions, no PM in the middle.
βββββββββββββββββββββββ
WHO YOU ARE
You're already living in this stuff. You read what comes out of Anthropic and OpenAI in real time, you've shipped with the APIs, you have opinions about which eval frameworks actually work and which are theater. This is the role you've been waiting for, not the role you applied to because the comp was good.
You move fast, but you've been burned enough times to know when fast is reckless. You can hold that trade-off.
You want your name on something real, not be engineer number 47 on someone else's feature team.
βββββββββββββββββββββββ
NICE-TO-HAVE
β’ Recruiting, HR-tech, or staffing-software experience
β’ Hired and led small engineering teams before
β’ Voice, telephony, or RTC experience
β’ Strong opinions about agent UX (what a tool should and shouldn't expose to an LLM)
βββββββββββββββββββββββ
WHAT YOU GET
β’ Greenfield, full ownership. You pick the architecture, you pick the tools. No legacy code, no committee, no permission slip.
β’ A founder who ships. Bernd (founder) uses Claude Code daily and builds production tooling himself. You get a counterpart who understands what you're doing, not someone managing you from outside.
β’ Above-market pay. We won't quote a band on the public posting, but we definitely pay properly for this role.
β’ The opportunity to take full responsibility of the system architecture short-term and build a team to support you long-term.
β’ Tooling and hardware budget that respects your craft. Premium AI subscriptions, top-spec machine, conference and training budget, trips to the Bay Area when there's a reason.
β’ Optional two retreats a year. Austria and Dubai. Proper time together with the founder and core team, not Zoom.
βββββββββββββββββββββββ
HIRING PROCESS
1. 30-min intro call with Bernd (founder)
2. 60-min architecture conversation, no leetcode, never
3. Paid pair-programming or take-home on a real Riffbird problem (4 to 8h, paid at market rate)
4. References, decision, offer within 7 days
βββββββββββββββββββββββ