Software Engineer, Backend Developer, AI Engineer (Applied​/Software)

Octocom is an AI agent company. We build AI agents that take real work off companies’ plates, starting with the front line: customer support, sales, and everything around them across email, chat, phone, and social. We already do this across verticals: e-commerce, telecom, insurance, healthcare, and pretty much anyone selling direct to consumers. If a company talks to a lot of customers, that’s our problem to solve.

Here’s the thing though. Wiring up an AI loop isn’t the hard part, anyone can do that now. The hard part, the part that actually creates value, is building a system that stays continuously in sync with a business’s real processes, systems, and data, and integrates with all of it automatically. No consultants, no dedicated IT project, no six‑month rollout. The agent just plugs in and works.

That’s the software of the future, and it’s what we’re building. Not a basic tool you bolt on, a genuine power‑user platform that solves problems companies actually have.

And it’s working. A few million in ARR, profitable, ~20% month‑on‑month growth for 18 months straight and not slowing down. We fund growth from revenue, so there’s no board dictating priorities and no all‑or‑nothing bets. Just a business that compounds, built by a small team that ships fast.

That’s the opportunity.

A small team runs the entire thing, so “full stack” here means the actual full stack: the AI agents themselves, the systems that keep them in sync with each business’s data and processes, and the interfaces customers and companies touch. Roughly 60-70% of the work is genuinely novel applied‑AI and systems engineering, not CRUD. The brief is usually “here’s a customer problem, go solve it,” and you own it from the design through the infra it runs on.

To make it concrete, here’s real work already in the codebase, the kind of thing you’d be building and extending:

• Novel AI agent architectures. Not “call an LLM and hope.” A real agentic loop with two‑phase model fallback across providers, workflow‑gated tool unlocking, and composable validators that catch a hallucinated order e it ever reaches a customer. An LLM‑judge that grades finished conversations against versioned checks. We even let an LLM write its own Python parsers and condition logic and run them in sandboxed cloud sessions.
• Real‑time voice. A phone agent that transcribes you live, thinks, and talks back with almost no lag. Turn‑taking and barge‑in state machines, speculative response generation that starts the moment it thinks you’re done, prompt‑cache warmup, latency traced per turn down to the millisecond. Three streaming providers (speech‑to‑text, LLM, text‑to‑speech) orchestrated in soft real time over the phone network. This is hard engineering and it’s a lot of fun.
• Systems at scale. Syncing product catalogs in the millions with streaming pipelines, Redis resume cursors, and content‑hash write‑elision so unchanged rows cost nothing. A custom job engine on Node worker threads with per‑isolate heap watchdogs and Redis‑persisted queues that keeps every conversation moving in parallel, reliably, rid product search (keyword + vector + exact‑SKU, rank‑fused) with a self‑tuning eval harness.
• Copilots and MCP as a real control plane. MCP isn’t a demo here, it’s how we actually drive production: thousands of lines exposing bots, workflows, search tuning, and read‑only SQL, all permission‑scoped and audited. Plus an internal copilot with web tools and a subagent that fans out work in parallel.
• Everything around it. React / Tan Stack / Tailwind dashboards. Integrations with ~55 platforms (Shopify, Zendesk, Gorgias, Magento, and the long tail). Email, SMS, chat, phone, and social channels. And yes, the occasional Saturday incident, because the best customers don’t take the weekend off either.

You won’t do all of this in your first month, and no one is a specialist in all of it. The point is the variety and the ownership: on a team this small, the interesting problems land on your desk instead of a 20‑person org’s backlog.

The tech:
Type Script end to end, React, Tan Stack Router, Tailwind, tRPC, PostgreSQL, Redis, worker threads, a vector store, sandboxed…