Senior Python Backend & AI/LLM Engineer

Senior / Staff Python Backend & AI/LLM Engineer Architecture, Applied AI, and Agentic Development Hands-on senior+ IC role | 70% engineering focus | Python, AI systems, architecture, and ADLC Company Description SANNOS is an intelligence layer for global compliance, providing an automated assessment and risk generation engine trusted by Big 4 firms and major institutions to manage complex regulatory environments. Founded by risk and compliance experts Ralph Bengtsson and Anders Søborg, the company was created from real-world experience on the frontlines of global regulation. SANNOS bridges high-stakes executive decision-making with advanced computing to automate audit and compliance reviews in minutes instead of days. The platform helps organizations turn compliance from a bureaucratic burden into a strategic advantage, supporting frameworks such as ISO27001, DORA, NIS2, NIST, CMMC, Secure Control Framework (SCF), and CSRD. Team members join a mission-driven environment focused on clarity, confidence, and innovation in compliance technology. The Role We’re looking for a Senior Python Backend & AI/LLM Engineer who can build production systems, make sound architecture decisions, and create an agentic developer platform or a smaller version of one. This is not a ticket-only backend role. You’ll own problems from design through production and help build reusable agent workflows, runtime integrations, safe updates, evaluation, and observability. You should use coding agents and LLMs throughout the Agentic Development Lifecycle (ADLC) while taking full responsibility for quality, security, and technical decisions. You should be able to show what you personally built, why you designed it that way, how it performed in production, and what business or engineering result improved. This is posted as a band. Whether you join at Senior or Staff level is decided at offer, based on the interview loop and the scope you have owned before — not by the title on this post. What You Will Do Spend about 70% of your time on hands-on engineering, including Python code, system design, AI workflows, debugging, and technical reviews. Design and operate Python APIs and services for users, devices, access rights, workflow catalogs, versioned releases, and telemetry. Package reusable AI workflows using skills, specialized agents, hooks, rules, tools, and supporting files. Build adapter layers that deliver the same workflows to different coding-agent runtimes and handle capability gaps safely. Orchestrate sequential and parallel agent tasks with shared state, isolation, retries, checkpoints, resumable work, and human approval. Build safe lifecycle operations with previews, file ownership tracking, conflict detection, snapshots, updates, and rollback. Create project context, agent instructions, tool contracts, and evaluation checks that produce consistent engineering results. Track workflow success, latency, failures, token usage, cost, and developer time saved; use the data to improve the system. Protect secrets and user data through safe defaults, permission boundaries, redaction, audit logs, and review gates. Use AI daily across ADLC and raise the team’s level through architecture reviews, code reviews, mentoring, and practical standards. Engage on architectual decisions Your Experience - Mandatory: 8+ years in software engineering, with direct ownership of production backend systems. 7+ years building Python services at production scale, using FastAPI, Django, Flask, or similar frameworks. 3+ years working with LLMs or coding agents in real projects, with at least 2 of those running in production - agent instructions, tool use, context design, and evaluations. Applied LLM engineering is a young field, so we weigh what you shipped more heavily than how long you have been doing it. Hands-on experience orchestrating multi-step or multi-agent workflows with shared state, parallel work, retries, failure recovery, and human approval. At least one tool, service, or internal platform you personally designed, built, and released — and can walk us through end to end. Strong fundamentals in API design, async work, idempotency, databases, object storage, queues, testing, cloud infrastructure, CI/CD, and observability. Practical security judgment for AI systems: secrets handling, prompt injection, data exposure, permission boundaries, model limits, token cost, and unsafe tool use. Evidence We Are Looking For For the AI part of this role, evidence matters more than years. This is how we actually assess. Be ready to walk through examples with clear results, such as: A tool, service, or internal platform you personally built and released, including users, scope, workflows, integrations, or scale. A multi-step or multi-agent workflow you shipped, including state management, failure recovery, and human review. A measurable result such as faster delivery, safer changes, lower token cost, higher success rate, or fewer failures. A case where you rejected or corrected AI-generated code or architecture because it was weak, unsafe, or incorrect. Your own code, architecture decisions, production lessons, and impact on the wider team. What We Are NOT Looking For A ticket-only developer whose main goal is to complete assigned tasks and log hours. Someone limited to basic CRUD work who does not own architecture, production behavior, or technical outcomes. Someone who calls an LLM API but has not built, evaluated, or operated a real AI feature. Someone who accepts AI-generated code or architecture without technical review, testing, and production accountability. Nice To Have Experience building cross-platform CLIs, local developer tools, desktop-app backends, or plugin marketplaces. Experience with agent frameworks, tool calling, MCP, hooks, event-driven workflows, or multi-agent collaboration. Experience with software licensing, device activation, billing entitlements, artifact registries, or signed releases. Experience leading technical discovery, platform architecture, security reviews, or engineering standards across teams. Experience building APIs for authentication, access rights, devices, versioned artifacts, release channels, and secure downloads. Experience with package or plugin lifecycle design, including install, update, compatibility checks, migration, and rollback. Experience building adapter or integration layers for external runtimes, CLIs, APIs, or tools with different capabilities. Experience designing safe local or remote changes using idempotency, dry-run previews, ownership checks, snapshots, and audit records. First 6-Month Success Measures - Our Expectations Exact targets will depend on the current product and baseline. Expected outcomes may include: Platform MVP: Support at least 1 coding runtime and 3 reusable engineering workflows Lifecycle: Provide safe install, update, conflict handling, snapshot, and rollback AI quality: Add automated evaluations for every critical agent workflow Operations: Track success, latency, failures, token use, cost, and time saved Safety: Add secret scanning and approval gates for all high-risk actions Reliability: Meet agreed service SLOs, including 99.9% availability where required Team growth: Ship 2 reusable ADLC patterns and run 1 practical session per quarter These numbers are working targets, not vanity metrics. The goal is to deliver reliable systems, useful AI, faster engineering, and clear technical ownership.