SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
bunch is building the operating infrastructure for private markets, combining AI-powered automation with regulatory expertise to replace fragmented processes with an integrated platform. The company has 4x ARR in 2025, crossed 150 fund managers and 12,000 LPs, and closed a $35M Series B in May 2026.
As a Senior Applied AI Engineer on the AI Platform team, you will take AI from working to relied upon. bunch already runs AI in production—a document extraction pipeline live in fund operations, agent workflows on Mastra, evaluations, and multi-provider fallback inside EU data residency. Fund operations depend on documents, deadlines, and numbers that must be correct: subscription documents to parse, capital calls to chase, portfolio data to reconcile.
You will own end-to-end product engineering: architecture, evaluation, integration, and deployment. Your top priorities include:
• Build and ship agents that automate real fund-operations workflows, integrated with services, data model, and authorization system—not standalone prototypes.
• Evaluate and improve agent performance through test cases built from real documents, regression suites in CI, human review where correctness is non-negotiable, and clear success criteria.
• Own the AI application architecture: orchestration and multi-agent design, tool contracts, memory, context engineering (RAG, MCP), plus guardrails, approvals, and human-in-the-loop controls for anything touching investor money.
• Run it in production: versioned, feature-flagged rollout; rate limits, provider fallback, and regional failover within EU data residency; observability to trace failures across services.
• Make it a platform: consolidate document parsing and extraction into this team; make shared evaluation and observability something other teams consume; set patterns other engineers inherit; mentor across teams on agent and LLM practice.
In your first 90 days, you will review the MCP offering end to end and ship it to first customers with proper access boundaries, evaluation, and observability; stand up shared observability for AI workloads (token usage, cost, latency, failures, retries) on a dashboard people use during incidents; and publish v1 of agent patterns (domain boundaries, tool contracts, prompt and eval conventions) adopted by at least one team outside AI Platform.
Tech stack: TypeScript, Svelte, React, Node.js, Nest.js, MySQL, PostgreSQL, Mastra, Vercel ai-sdk, Google Vertex (Gemini) with AWS Bedrock fallback in EU regions, Kubernetes on AWS, Datadog, FusionAuth, Retool.
The role is hybrid (3 days/week in office in Berlin) with up to 6 remote calendar weeks per year. bunch has 130+ employees from 40+ countries.
REQUIREMENTS:
• 5+ years building production software, including at least one agent or LLM-powered capability taken end to end and owned post-launch
• Real depth in orchestration and context engineering: tool contracts, memory, RAG, multi-agent design with Mastra, ai-sdk, LangGraph, or similar; integrated agents into real product and authorization model, not prototypes
• Ability to make non-deterministic systems measurable: test cases from real data, regression suites, human review; distinguish between model improvement and benchmark drift
• Production ownership: reading traces, diagnosing failures, tuning cost and latency, handling provider rate limits and fallback
• Experience with TypeScript and/or Python
• Platform mindset: build for other engineers; standards adopted through trust, not documentation
• Pragmatic approach: start from business outcome, choose deterministic solutions when appropriate, know when good enough is sufficient
• Plus: fintech, private markets, or regulated document-heavy domain experience; experience with EU data-residency constraints
