AI Agentic Engineer

Profitmind is a retail analytics SaaS company that turns competitive and customer data into agent-driven insights, helping retailers make faster, sharper merchandising and pricing decisions. Based in Pittsburgh and backed by a recent strategic investment from Accenture, we're scaling our agentic AI platform for some of the world's largest retailers. Our platform runs a team of nine specialized agents — Data Load, Strategy, Competitive Intelligence, Pricing, Inventory, Promotions, Assortment, Planning, and the Monday Morning agent — that analyze a retailer's entire business every week and hand merchandising teams a ranked list of actions with the dollar value attached. This role builds the agentic stack that makes that possible. The RoleWe're hiring a multi-disciplinary AI Agentic Engineer to design, implement, and deploy LLM-driven agents with strong backend and front-end integration — across agent harness engineering, agent development, agentic workflows and platform integration, LLM inference/evaluation/hosting, and fine-tuning — building production-grade applications on top of all of it. The ideal candidate combines a strong Python and AI foundation with hands-on LLM knowledge (prompt engineering, context management, structured outputs and tool calling, retrieval, evals, and fine-tuning with LoRA/QLoRA), harnesses (Claude Agent SDK, OpenAI Codex SDK, Pi, OpenClaw, Hermes Agent), and agent frameworks (LangGraph, PydanticAI, Google ADK). What You'll DoAgent Harness Engineering • Build and own the harness layer — agent loops, tool execution, context management, session persistence, permission controls, and sandboxed environments — evaluating and extending existing harnesses where they beat building internally. • Architect orchestration for multi-step planning, long-term memory, and dynamic tool use, with guardrails, fallbacks, and self-correction for timeouts, context overflows, and hallucinated tool calls. • Standardize internal API contracts for tool creation so new enterprise data sources plug into the agent environment securely.Agent Development • Architect and build production-grade LLM agents (LangGraph, PydanticAI, Google ADK, or custom loops when a framework adds unnecessary complexity), with composable patterns for retrieval, planning, reflection, and subagent delegation. • Integrate vector databases and knowledge graphs for RAG, memory, and grounded decision-making. • Engineer context deliberately (just-in-time retrieval, progressive disclosure, compaction, structured note-taking, isolation) and evaluate prompt/tool strategies for reliability.Agentic Workflows and Platform Integration • Build multi-step, multi-agent workflows with routing, parallel execution, checkpointing, retries, compensation, and human approval steps that survive restarts and degrade gracefully. • Build MCP servers for structured, auditable, least-privilege access to internal systems, and scalable FastAPI services for synchronous, asynchronous, and streaming execution. • Connect agents to internal applications, chat platforms, triggers, and CI pipelines; build agent interfaces (React, TypeScript, Next.js) with real-time streaming (SSE/WebSockets) and clear UX for long-running agents — progress, interruption, steering, approval, recovery.LLM Inference, Evaluation, and Hosting • Evaluate and integrate managed or self-hosted models; monitor and improve quality, latency, reliability, and cost (caching, batching, model routing). • Build task-specific evals and regression tests using Profitmind's real retail workflows, and instrument agent behavior — tool calls, errors, latency, tokens, cost.LLM Fine-Tuning and Continuous Improvement • Use eval results to decide between prompting, context engineering, retrieval, or fine-tuning; run targeted LoRA/QLoRA experiments when justified. • Help curate and version evaluation and training data from representative business use cases. What We're Looking For • Education and foundation. Bachelor's or master's in computer science or related, or equivalent practical experience, with 2+ years building agentic applications. • Production agent experience. 1+ year operating LLM agents in production, with frameworks (LangGraph, PydanticAI) and SDKs/harnesses (Claude Agent SDK, OpenAI Codex SDK). • Agent architecture. Deep understanding of agent loops, tool calling, retrieval, and context management, with proven ability to measure and improve agent performance. • Evaluation and fine-tuning. Experience evaluating non-deterministic AI systems and fine-tuning or adapting models for domain-specific tasks. • Backend and data infrastructure. Production experience with FastAPI, Docker, MLOps practices, and vector stores. • AI-assisted engineering. Proficiency with agentic coding assistants (Claude Code, Codex, Cursor), folding new tooling into how you work rather than around it. Nice to Have • Hands-on fine-tuning (LoRA or QLoRA) and curating evaluation or training datasets • Knowledge graphs alongside vector re