Job description
Job Opportunity only available for professionals located in Latin America.
The Principal Engineer, AI Orchestration & Retrieval defines how the enterprise's central AI system is composed – the orchestration and abstraction layers that connect LLMs to tools, data, and one another, and the retrieval systems that ground them. This role sets the strategy and builds the reality for how we build and expose tools (including MCP servers), how we structure retrieval and chunking, and when to rely on specialized sub-agents versus directly exposing tools to a model.
Requirements:
Advanced English level (C1/C2) to ensure fluent communication across teams.
10 or more years of experience in software/AI engineering, with hands-on experience building LLM orchestration, agents, and retrieval systems.
+4/5 years of deep hands-on experience with LLM orchestration frameworks (e.g., LangGraph, LlamaIndex, Semantic Kernel, or equivalents) and agentic patterns.
Direct experience building MCP servers and tool/function-calling integrations.
Evidence-based opinions on the optimal number of tools to expose to an LLM and the optimal number of APIs per MCP server, and on overall tool-surface design.
A clear, defensible point of view on specialized sub-agents versus direct tool exposure, and the tradeoffs of each.
Deep experience with retrieval/RAG: chunking strategies, embeddings, vector databases, hybrid search, and re-ranking.
Experience designing abstraction layers and platform APIs that many teams build on top of.
Strong understanding of context-window management, prompt/context assembly, and cost/latency optimization.
Experience with evaluation and observability for agentic and retrieval systems.
Ability to set strategy and standards while remaining hands-on in code.
Strong stakeholder collaboration and problem-solving skills.
Essential Responsibilities:
Design the orchestration and abstraction layers of the central AI system that connect LLMs to tools, data, and sub-agents.
Design, build, and operate MCP (Model Context Protocol) servers and set standards for how tools are defined, exposed, and versioned.
Define tool-surface strategy: the optimal number of tools exposed to an LLM, the optimal number of APIs per MCP server, and how to keep tool surfaces coherent and discoverable.
Establish when to use specialized sub-agents versus directly exposing tools to a model, and design the corresponding multi-agent patterns.
Design retrieval (RAG) systems: chunking strategies, embedding models, vector stores, hybrid/keyword search, re-ranking, and context assembly.
Define abstraction layers that decouple product teams from the underlying models, tools, and providers.
Build routing, context-window management, and memory strategies for agentic workflows.
Define evaluation for orchestration and retrieval quality (retrieval precision/recall, tool-selection accuracy, task success, latency, and cost).
Establish observability and tracing across multi-step agent and tool calls.
Address safety, guardrails, authentication, and access control across tools and agents.
Partner with product teams to onboard their capabilities as tools and agents into the central AI system.
Mentor engineers and raise orchestration and retrieval maturity across teams.
What do we offer?
Competitive salary in USD.
100% Remote Work.
Type of contract: Independent Contractor with Venon Solutions LLC.
Contract duration: long-term
Benefits: 2 weeks of PTO (paid time off).
Holidays: from the North American calendar.
Working hours: Full-time EST, fully committed.
Originally posted on Himalayas
Who can apply
Eligible countries: United States. Accepted UTC offsets: UTC-10, UTC-9, UTC-8, UTC-7, UTC-6, UTC-5, UTC+14. Review the full description for employer-specific work authorization, residency and schedule requirements.