Backend Engineer

About Conduit

AI customer service is one of the most funded categories in tech right now. The customer reviews for the biggest players average under 2 out of 5 stars. 75% of consumers leave AI support interactions frustrated. Companies report 90%+ CSAT scores while their customers rate them 40 points lower. That gap exists because nobody built the tools to see what's actually happening.

Conduit is building the infrastructure that makes AI customer service actually work. Not a chatbot. Not a deflection engine. A platform where every AI decision is traceable, every answer links back to a source, and the people who know the customer best (the CX team, not engineers) are the ones running the system.

We believe customer service is going through the same compression that happened in software engineering and GTM. What used to require a team of specialists will be done by one empowered operator. We call them conversation engineers. Conduit is the platform they use to do the job.

Role: Backend Engineer

The backend is the product. The reliability of message delivery, the speed of AI responses, the depth of integrations, the transparency of AI decision-making, customers feel all of it even if they never see your code.

You'll work directly with Punn (CTO) and a small engineering team where your commits hit production the same day and customers notice.

Why this role is different

• You're building the observability layer for AI. Every other platform in this space is a black box. We're building the cockpit: structured traces of what the AI read, what it ignored, where confidence dropped, and why it chose what it chose. This is a hard, unsolved infrastructure problem and it's the core of what makes Conduit defensible.
• You're building the action layer, not just the answer layer. Most AI support platforms stop at information retrieval. Our AI agents execute multi-step workflows: modify a reservation, assign a maintenance request, text the vendor, confirm back to the customer, close the loop. You own the workflow engine that makes all of that reliable.
• You're solving a dirty data problem. Enterprise data is messy, incomplete, contradictory, and constantly changing. The AI doesn't hallucinate, it faithfully grounds answers in whatever broken context it's given. You're building the systems that surface what's missing, what's stale, and what caused a wrong answer before 10,000 customers are affected.
• You're building for a new type of user. Our platform is designed for CX operators, not developers. That means your backend has to be rock solid because there's no engineer on the customer's side to work around your bugs. The abstractions you build are the product.

What the work looks like

• Real-time messaging infrastructure. SMS, WhatsApp, email, voice. Webhook handlers, retries, delivery receipts, per-workspace rate limits. Sub-second delivery matters because the person on the other end is waiting.
• Workflow execution engine. Reliable step execution with retries and backoff. Time-based triggers. Pause/resume. Idempotency. These workflows are the "hands" of our AI agents.
• AI trace pipeline. Every AI response gets a full decision trace: sources retrieved, context assembled, confidence signals, fallback paths. This is how customers audit their AI and how we close the perception gap.
• Integration layer. Property management systems, CRMs, payment processors, and a growing list of systems our AI needs to act on. Timeouts, circuit breakers, dead letter queues. Third-party APIs are unreliable. Our system can't be.
• Data model evolution. Zero-downtime migrations, backfill scripts with verification and rollback. You'll be evolving schemas under a running system that handles real conversations.
• Omnichannel threading. Conversations that span calls, texts, emails, and chat, all stitched into one coherent thread per contact. Correct threading is deceptively hard and critically important.

Who you are

• You've shipped and operated production services. You debug with logs, metrics, and traces, not guesses.
• You default to the simple fix. You add complexity only when the system forces your hand.
• You write design docs and migration plans before you write code. You think about failure modes first.
• You want to own the full lifecycle. Design, ship, deploy, get paged at 3am, fix it, make sure it never happens again.
• You want a small team where you have real ownership and direct impact, not a cog-in-the-machine role at a company with 200 backend engineers.

Compensation & Benefits

• $150-250K base
• Ownership of critical infrastructure in a product that's growing fast
• Day-1 architectural influence
• Direct line to the CTO and the customer

Interview Process

• Initial Application: Short Video (1 minute): Introduce yourself and why you’re a strong fit.
• First Call (20 min) Fit assessment with the founders to understand your experience and goals.
• Take-home Project (2–4 hours) Implement a technically challenging backend project end-to-end.
• Assignment Discussion (45 min) Talk through your take-home project. Discuss potential solutions and tradeoffs.
• Technical Interview (45 min) A coding challenge focused on backend infrastructure or data pipeline scenarios.
• On-site Work Trial (1–2 days in San Francisco) Spend a few days with the team (travel covered), outlining a potential backend feature and deploying a simple version to production.
• References, Decision, & Offer

We'd love to hear from you if you’re passionate about building real-world AI that will transform communication for service-oriented businesses.