AI Engineer

Our Client builds production-grade AI sales agents for businesses in Southeast Asia, deployed on WhatsApp and Instagram. The agents run real sales conversations: answering questions, recommending products, taking bookings, processing payments, and following up automatically.

They are backed by Lightspeed Venture Partners, Monk's Hill Ventures, and M Venture Partners, with advisors including Kevin Aluwi (co-founder and ex-CEO of Gojek). It has 20+ enterprise clients in production.

Key Responsibilities

• The internal AI tooling. Agent builder, agent fixer, copilot analyzer — you build and improve these so the Forward Deployed Engineering (FDE) team can ship better agents faster.
• Harness and context engineering. You design and optimize the prompts, pipelines, and orchestration logic that make our agents actually perform in production.
• Evaluation and debugging pipelines. You build the systems that catch AI failures before clients do — automated testing, AI judges, regression detection.
• Model integration and routing. You stay current on what models are available, what they're good at, and you plug the right ones in at the right places.
• AI infrastructure layer. You work with AWS, you understand latency and cost tradeoffs, and you make smart decisions about what runs where.
• Shipping features end-to-end. You don't hand off to someone else. You design, build, test, and ship. You own it until it's in production and working.

Key Requirements

• Builds agents, not just chatbots. You understand the difference between a prompt and a system. You've built multi-step agent workflows, tool-calling pipelines, or orchestration logic — not just single-turn LLM calls. You think in context, retrieval, routing, and evaluation.
• Harness and context engineering depth. You know how to engineer a prompt system that's robust, not brittle. You understand context window management, structured outputs, chaining, and how to make an LLM do exactly what you need it to do even when the input is messy.
• Ships fast and ships clean. You use Claude Code, Cursor, or whatever it takes to move at speed. But you also know how to review AI-generated code critically — you don't just accept the output, you own it. Your code goes to production, not just to demo.
• Debugging instinct. When an agent fails, you don't shrug. You trace the failure, identify the root cause — whether it's the prompt, the context, the model, or the pipeline — and you fix it. You're the person who builds the tools that catch failures before they reach the client.
• AWS and server literacy. Not DevOps-level, but you're comfortable with EC2, understand how to monitor a service, read logs when something breaks, and think about cost when you design something. You don't ship something that will fall over at 10x load.
• Product sense. You understand who uses what you build. When you design the agent fixer or the copilot analyzer, you think about what the FDE actually needs — not just what's technically elegant. You ask "does this make their job meaningfully easier" before you ship.
• Full ownership of your scope. You don't wait for a ticket to tell you something is broken. You don't finish a feature and walk away. You monitor, you iterate, you improve. Your KPI is whether the thing you built actually works in production, not whether the PR got merged.

Nice-to-haves

• Experience with LLM evaluation frameworks (Langsmith, RAGAS, or custom eval pipelines)
• Has built internal tooling used by non-technical teams (ops, sales, support)
• Familiarity with RAG architecture, vector databases, or retrieval pipelines
• Experience with multiple LLM providers and their tradeoffs (OpenAI, Anthropic, Gemini)
• Has worked in a sales, CRM, or customer-facing AI product before