Ai Software Engineer

Build the multi-agent core of an AI-native TMS for freight & logistics. FreighAi is an AI-native TMS (Transport Management System) for freight forwarders and logistics businesses — automating day-to-day operations at roughly 80% agentic AI, 20% human-in-the-loop. The product is live and serving customers today. We’re hiring AI Software Engineers to extend it — build advanced versions, ship enhancements, maintain and harden what’s running, and troubleshoot production issues across our multi-agent systems and integrations. OUR STACK

• Backend: Microservices architecture — Python (primary); Kotlin / Java a plus
• Frontend: React.js
• Database: MongoDB
• AI: A mix of LLMs and SLMs, multi-agent orchestration, RAG + vector search HOW WE BUILD The majority of development happens in Claude Code, with agentic coding as the default workflow. We work across a mix of LLMs and SLMs — picking the right model for each job (large models for reasoning and orchestration, smaller / faster models for high-volume, latency- or cost-sensitive tasks). You’ll be orchestrating agents and reviewing their output as much as writing code by hand. This is a build-and-own role, not a research role — you’ll be shipping to real freight customers quickly. WHAT YOU’LL DO
• Extend and enhance our multi-agent workflows (planner, executor, reviewer, and domain-specific agents) handling real freight ops — quoting, booking, track-and-trace, document processing, exception handling.
• Maintain, troubleshoot, and harden production systems already serving customers — diagnose agent failures, fix regressions, improve reliability and cost.
• Design and own services in our microservices backend — clean APIs, sensible boundaries, resilient inter-service communication.
• Own integrations: FreighAi works as a TMS and integrates with other TMSes, plus carrier / rate APIs (ocean, air, road), ERPs, EDI (204/210/214/990, etc.), customs / compliance platforms, and email / document ingestion.
• Build and improve document AI pipelines (OCR + extraction) for BOLs, invoices, packing lists, and customs paperwork.
• Build and maintain clean, responsive React.js front-ends for the operators who handle the human-in-the-loop side.
• Tune model selection and routing across LLMs/SLMs, and refine prompts, tools, memory, and retrieval (RAG + vector search) for reliability and unit economics.
• Strengthen human-in-the-loop flows — escalation, approvals, confidence thresholds — so humans handle the ~20% cleanly.
• Improve evals, observability, and guardrails — catching hallucinations, retries / fallbacks, latency / cost monitoring, and tracking agent success rates in production. WHAT WE’RE LOOKING FOR
• 2–5 years of professional software engineering experience, with real production systems under your belt.
• Strong backend engineering in Python within a microservices architecture — async, queues, API design, service boundaries — and the judgment to work on a live system without breaking it. Kotlin / Java experience a plus.
• Solid React.js front-end skills — you can build and maintain production UI, not just prototypes.
• Strong MongoDB skills — schema design, aggregation pipelines, indexing, and query performance on production data.
• Hands-on experience building LLM-powered or agentic applications: tool use, orchestration, prompt design, evals, RAG (LangGraph, CrewAI, AutoGen, MCP, or custom).
• Fluency with agentic coding tools (Claude Code, Codex, or similar) as a primary way of working — not just occasional autocomplete.
• Experience with messy third-party API integrations; bonus for EDI, TMS, or any logistics / supply-chain exposure.
• Strong English communication — written and verbal. You can write clear specs, explain trade-offs, and collaborate with the team and customers. This is non-negotiable: clear writing is how you delegate to agents and how you’ll work here day to day.
• A production-reliability mindset — you treat agents as systems to be monitored, tested, and hardened, not demos. THE AI-FIRST WORKING STYLE WE WANT
• Clear written communication and spec-writing — you can turn a fuzzy requirement into a precise brief an agent can execute. Clarity of intent is your core skill.
• Context engineering — you know what to feed the agent (conventions, constraints, relevant files) and what to leave out; you keep project context sharp.
• Task decomposition — you break work into agent-sized chunks with clear, verifiable done-criteria.
• Critical review at scale — you read AI-generated diffs skeptically, catch subtle bugs and bad patterns, and never rubber-stamp.
• Verification discipline — you think Plan → Execute → Verify: tests and specs first, validation always.
• Good judgment on when to drop down — you know when to let the agent run vs. when to hand-write the tricky bit yourself. NICE TO HAVE
• Freight forwarding, supply chain, or logistics domain knowledge.
• Vector databases, document AI / OCR, and RAG at scale.
• Prior work on human-in-the-loop or workflow-automation products.