Senior AI Engineer | Coding Agents, AI Generated Slides, Python-pptx knowledge

We are a product company building a production-grade multi-agent system powered by RAG and LangGraph. One of our key agents will generate PowerPoint presentations from natural language by writing code that produces a PPTX file. We are looking for an engineer who has real experience building coding agents, not just prompt writing. You should be comfortable turning user intent into structured code, validating outputs, handling failures safely, and integrating the agent into a broader orchestration layer. Keep it short and to-point, we are looking to identify and hire quickly. A key requirement for our decks is charting. We rely on think-cell for executive-quality charts and we need the agent to create and populate think-cell charts in PowerPoint in a reliable, repeatable way. WHAT YOU WILL BUILD Phase 1: Template-driven generation (reliable, brand-safe) - Convert user instructions into slide plans based on predefined layouts and components - Generate PowerPoint via code (PPTX generation libraries) reliably and consistently - Enforce constraints like brand styles, spacing rules, typography, and layout limits - Create and populate think-cell charts using our templates and data contracts - Examples: bar, column, waterfall, Mekko, line, area, stacked, Gantt, and other think-cell chart types we standardize on - Ensure the output opens cleanly in PowerPoint with think-cell installed and renders correctly Phase 2: Freeform generation (more flexible, still high quality) - Expand from fixed layouts to flexible slide creation from user descriptions - Handle more complex requests like diagrams, tables, multi-column layouts, icons, and narrative flow - Create advanced charting flows where the agent selects chart types, shapes the story, and drives think-cell chart data - Build guardrails so creative freedom does not reduce quality or reliability RESPONSIBILITIES - Design and implement a coding agent that: - Interprets natural language into a structured slide specification (a slide plan and a chart plan) - Produces code that generates PPTX presentations from that specification - Iterates when output quality fails checks (layout overflow, missing assets, broken formatting, chart rendering issues) - Build chart generation that works with think-cell: - Use template-based chart placeholders and fill the underlying chart data through supported automation approaches - Keep chart styling consistent with our standards (colors, labels, typography, spacing) - Validate chart correctness (data mapping, labels, totals, axis logic, and expected chart type behavior) - Integrate the agent into our multi-agent system: - LangGraph workflows, tool use, memory patterns, and state management - MCP-based tool layers and integration patterns (tool protocol design, tool routing, structured tool contracts) - Build a production-grade execution layer: - Sandboxed code execution or constrained generation approaches - Deterministic rendering where possible, with clear fallbacks - Observability: logs, traces, and structured debugging artifacts (including slide specs and chart specs) - Add quality controls: - Automated checks for slide validity and layout compliance - Automated checks for chart validity and think-cell compatibility - Regression tests and evaluation harnesses for common prompt types and common deck patterns - Collaborate with product and engineering to evolve capabilities over time CORE REQUIREMENTS - Strong software engineering fundamentals with Python or JavaScript/TypeScript - Hands-on experience building coding agents that generate or modify code from natural language - Experience with agent frameworks and orchestration concepts (LangGraph preferred) - Familiarity with MCP concepts (tool contracts, tool routing, structured outputs, and safe tool execution) - Practical backend integration experience: - APIs using FastAPI and/or Node.js - Async workflows, queues, retries, idempotency, and long-running jobs - Experience generating PowerPoint via code: - Examples include PptxGenJS, python-pptx, or Open XML based generation - Ability to debug the generated PPTX and understand common failure modes - Experience with think-cell chart workflows in PowerPoint: - Strong practical understanding of how think-cell charts are created, templated, and data-populated - Ability to design an automation approach that is stable in production and easy to test - Strong understanding of the PPTX file format and related XML (OOXML) - Experience generating PDFs or other document formats from code NICE TO HAVE - Experience with cloud deployment, CI/CD, and observability tooling - Experience with code-capable LLMs (for example GPT-5.2 Codex or equivalent) in real systems - Security mindset for code generation workflows (sandboxing, policy checks, safe execution) WHAT TO INCLUDE IN YOUR APPLICATION - A short note on a coding agent you have built: - What it generated, how you ensured correctness, and what failed in real usage - 1 to 2 examples of relevant work: - GitHub links, demos, or a clear architecture summary if code is private - Your availability and timezone - Confirmation of full-time remote interest - If you have done chart automation: - Describe how you created and validated charts (especially think-cell) and how you handled edge cases ENGAGEMENT - Remote - Full-time preferred - Strong communication and ownership mindset required INTERVIEW PROCESS - Short call on your past systems and agent building approach - Practical scenario discussion: turning a real prompt into a slide specification and a code strategy - Final conversation on delivery style, collaboration, and ownership