Hiring: AI Agent Workflow Engineer to help build an open-source self-hosted Voice Agent OS

We’re building Feros: an open-source, self-hosted Voice Agent OS for production voice AI.

The stack already includes:

  • a Rust real-time voice runtime
  • an AI-powered builder that turns natural language into agent graphs
  • integrations, OAuth/credential management, and custom webhook support
  • text debugging, auto-testing, versioning, and deployment

We’re now looking for an engineer who is strong in n8n and AI agent workflow design to help us make n8n a first-class
What you’d work on:

  • n8n-based workflow orchestration for voice agents
  • webhook/API/OAuth integrations between agents and business systems
  • reusable workflow templates for CRM, scheduling, lead capture, and support use cases
  • improving agent flow across prompts, tool calls, external actions, and evaluation loops
  • helping shape the open-source integration layer around the project

What we’re looking for:

  • strong hands-on n8n experience
  • solid understanding of AI agent workflows, tool calling, and automation design
  • comfortable with Python
  • able to work across APIs, auth, integrations, and product logic
  • TypeScript/React is a plus
  • Rust is a plus, but not required

This is a good fit if you like building at the boundary of AI agents, workflow automation, and production
infrastructure.

Repo: GitHub - ferosai/feros: Open-source voice agent OS. Rust runtime, AI-driven builder, sub second latency. Self-host everything. · GitHub

Interested? Join our team Discord: feros - Open Voice Agent OS and DM admin

2 Likes

Hi @dave_zhou
n8n workflow design and AI agent orchestration is exactly what I do. This looks like a genuinely exciting project… sent you a DM

@dave_zhou

Hi. I sent you a dm. My nickname is Nicolay_Pilguy

I have 3+ years of experience building advanced automation and AI-driven workflows using n8n, including API integrations, webhook-based systems, and designing scalable agent-like workflows with structured logic and external tool interactions.

I’ve worked on building modular workflows for CRM, lead management, and automation systems, and I’m comfortable designing flows that connect prompts, actions, and external services in a reliable way.

I’m also a top supporter of the n8n community and enjoy working on systems that sit at the intersection of AI agents and automation infrastructure.

I’d love to contribute and help shape the n8n integration layer for Feros.

Happy to connect and discuss further,Sent You dm.

You do not need another automation person who can wire up a few nodes and call it AI infrastructure. You need someone who understands where orchestration breaks when real agent state, auth, tool usage, webhook behavior, and production reliability all start colliding. That is the part I am strong in, and it is why I am a serious fit for Feros.

What makes this interesting is that the hard technical foundation is already there. The weak point in products like this is usually the layer between agent intent and business execution. That is where systems get messy, templates become one-offs, OAuth turns fragile, and evaluation stops being meaningful. I am good at stepping into that middle layer, structuring it properly, and making it usable by both the product and the engineering team.

  • I would start by reviewing where orchestration responsibility should live across the runtime, the graph layer, and n8n, because that decision will drive how clean or messy everything becomes afterward.

  • I would inspect how agent state is currently passed through prompts, tool calls, and external actions so n8n workflows receive enough structured context to make correct downstream decisions.

  • Before building templates, I would identify the highest value workflow patterns to standardize first, especially the ones most likely to expose weaknesses in auth, retries, data mapping, and state transitions.

  • I would define a repeatable contract for webhook and API interactions so you do not end up with a pile of custom logic that works once and becomes painful to maintain.

  • I would evaluate how OAuth and credential handling are currently modeled, because reusable workflow templates fail fast if credentials are not scoped, refreshed, and injected cleanly.

  • I would structure templates around real operational paths, not demo cases, so CRM updates, scheduling, support actions, and lead capture flows can actually be reused across agents.

  • I would review how tool calling currently behaves when external systems return slow responses, partial failures, invalid payloads, or conflicting records, because that is where production voice systems usually start showing cracks.

  • I would make sure evaluation loops measure whether a workflow produced the correct business outcome, not just whether a node ran without throwing an error.

  • I would look closely at how versioning should work between agent graphs and workflow assets so updates do not quietly break existing deployments.

  • I would help shape how n8n surfaces inside the product so it feels native to the agent system instead of looking like a separate bolt-on that users have to mentally translate.

  • I would keep the architecture pragmatic by pushing workflow logic into n8n where flexibility matters and keeping validation, adapters, and critical reliability paths in code where they belong.

  • I would also pressure test the integration layer for open source growth, because once outside contributors get involved, unclear conventions become a liability fast.

A few examples that are directly relevant:

Cococure AI WhatsApp Chatbot
This project is relevant because it involved building AI driven conversation flows that had to trigger the right business actions under real operational conditions, not sandbox logic. I served as Technical Consultant, Product Owner, and Project Manager, where I defined the orchestration structure, broke down the implementation into milestones, shaped how prompts interacted with external availability data, oversaw QA, and made sure the system behaved reliably once connected to live business workflows. The stack included OpenAI, LangChain, FAISS, Redis, FastAPI, WATI, and live API integrations. The part that ties most directly to Feros is that I was not just managing a chatbot. I was shaping the logic between AI behavior, external tools, system actions, and repeatable outcomes.

Select Screening Services Platform (https://stage.drugscreening-fe.testyourapp.online/)
This is relevant because it required building structured multi-role workflows where actions, statuses, external integrations, and audit trails all had to stay coherent across a live platform. I led the platform planning, technical direction, API structure, role based behavior, integration strategy, and delivery sequencing. That included handling where state changes should occur, how downstream actions should be triggered, and how to keep workflows understandable instead of letting them sprawl. The connection to Feros is direct: when you are building an orchestration layer for agents, clarity around system boundaries, execution order, and reliability matters more than clever demos.

YacDaddy.com SaaS Platform
This platform mattered because it replaced fragmented tools with a unified system that had to support subscriptions, role based actions, feature gating, operational workflows, and external integrations without becoming brittle. I led product and delivery planning, defined how moving parts should connect, and kept the implementation grounded in how the business actually operates. What ties this to your project is the need to turn complexity into a reusable system rather than a pile of special cases. That is exactly the challenge with building first class workflow support around AI agents.

I am open to a call if you want to get specific about how you want n8n represented inside Feros and where you see the current orchestration gaps.

  • Where do you want execution authority to sit when agent logic and workflow logic overlap or conflict?

  • Which use case do you see as the best first proving ground for template design: CRM, scheduling, support, or lead capture?

  • How mature is the current OAuth and credential model for tenant level or environment level reuse?

  • Are you treating n8n as an external orchestration engine, or do you want it exposed as a native part of the builder and graph system?

  • How are you currently measuring whether an agent workflow succeeded in a way that matters beyond technical completion?

Brandon

[email protected]

Hi Dave,

Voice AI + n8n workflow orchestration is my core stack — I’ve shipped a production voice agent for a healthcare client that handles real inbound patient calls, manages appointment scheduling, answers FAQs, and routes complex cases to human staff. Live, handling real callers daily. More on this and other projects: priyanshukumar.co

On the workflow orchestration side: I’ve built a multi-agent system where an orchestrator routes tasks to specialist agents based on intent classification, each agent has access to specific tool sets (web search, database queries, code execution, browser automation), and results flow through a QA review before delivery.

The Feros architecture — reusable workflow templates for CRM/scheduling/lead capture, agent flow across prompts, tool calls, and evaluation loops — maps almost exactly to what I’ve already built and deployed.

Stack: Python, n8n, Claude API, Supabase, FastAPI, Docker. Comfortable with TypeScript/React for frontend components.

Would love to contribute — happy to start with a specific workflow template (e.g., the CRM/scheduling integration) and expand from there.

Priyanshu Kumar
AI & Automation Engineer
priyanshukumar.co