Senior Applied AI Engineer

About Triquetra Health Triquetra Health is a premium dietary supplement sold across Amazon, TikTok Shop, our DTC store, and retail (GNC, Sprouts). We are a team of 70+ and have been in business for over 13 years. We're not a company "adopting AI." We already run AI in daily operations across product development, marketing, compliance, and operations — 60+ custom Claude skills, multi-agent orchestration systems, and a range of functional AI tools that need further refinement and true deployment to be fully usable across the team. Long-term, this consolidates into a unified internal platform we call Triquetra OS. Our subject matter experts use Claude Code and the Anthropic platform every day to build real systems that move the business. We're hiring you to scale and harden what we've already built, and to be the engineer who takes our team's prototypes the last mile. The Role This is a hands-on senior engineering role with high autonomy. You're not designing our AI strategy — our SMEs and CEO own that. You're the one who takes their work and turns it into reliable, multi-user systems the entire team can rely on. Subject matter experts across the company — marketing, R&D, operations, leadership — design AI workflows and prototype core functionality with Claude Code. You take those prototypes the last mile: hardening them, hosting them, building the interfaces, deploying them, and making them available and reliable to the broader team. Over time, you help consolidate ad-hoc internal tools into a unified internal platform. You'll work most closely with the CEO and with marketing, R&D, and operations stakeholders. Think of yourself as a shared service and force multiplier — your week might span finishing a TikTok/Reddit market research tool for the marketing team, deploying a social media creative asset generator for ambassadors, productionizing an R&D documentation skill, and quietly consolidating four one-off tools into something the team can navigate from a single login. How This Role Works in Practice

• You inherit prototypes, not blank pages. Stakeholders bring you working logic, clear specs, and a definition of "good." Sometimes it's a Claude Code project that runs locally for one person. Sometimes it's a half-built tool that never made it to deployment. Your job is to get it to the finish line.
• You translate ambiguous handoffs into shipped systems. Specs from non-technical SMEs won't always be complete. You ask the right questions, fill in the gaps, and ship. You don't wait for a perfect requirements doc.
• You orchestrate agents. A meaningful share of what you'll build involves multi-agent and multi-skill orchestration — chaining Claude calls, MCP servers, tool use, and external APIs into reliable end-to-end workflows.
• You own reliability. If a tool you built breaks for someone on the team at 11 PM, you know about it before they do. You've built the monitoring and evals to catch drift, hallucination, or degraded outputs before they impact the business.
• You consolidate over time. Individual tools may launch as standalone apps with their own logins. That's fine. But you're always thinking about how things converge into a unified internal platform, and you push that consolidation when it makes sense.
• You may prototype product-side AI features. You'll occasionally build core, functional prototypes for customer-facing AI on our broader platform vision (patient portal, AI personalization, Expert Network). The polish, hardening, and ongoing management of finished customer-facing product will lean on external developers - your job there is the working prototype, not the production product.

Responsibilities

• Take partially-built AI prototypes from stakeholders and ship them as production-grade, multi-user internal tools
• Design, build, and orchestrate multi-agent and multi-skill workflows using Claude, MCP, and tool use
• Build and own internal AI tools used daily by staff — intake assistants, review helpers, knowledge tools, automation pipelines, content generation systems
• Establish monitoring, evals, and validation frameworks to catch model drift, hallucination, or degraded outputs before they reach the team
• Manage prompt versioning, skill versioning, documentation, and change controls
• Integrate third-party APIs and data sources (TikTok, Reddit, Apify, Amazon, internal data) into AI

workflows reliably and securely

• Stay on the bleeding edge of Anthropic and other AI releases, frontier model capabilities, and the broader AI tooling ecosystem; bring new capabilities into our stack proactively
• Help consolidate ad-hoc internal tools into a unified internal platform ("Triquetra OS") over time
• Build core, functional prototypes for customer-facing AI features on our platform vision; collaborate with external developers on production hardening

Required Tools & Technical Skills Anthropic & Claude

• Anthropic Claude API — deep familiarity with system prompts, context window management, multi-turn design, and prompt caching
• Claude tool use / function calling — building reliable tool-augmented workflows
• Model Context Protocol (MCP) — connecting Claude to external tools, APIs, and data sources; building or extending MCP servers
• Claude Skills — designing, building, and maintaining reusable skill packages
• Claude-specific prompt engineering — XML structuring, extended thinking, instruction hierarchy, output formatting
• RAG implementation with Claude — retrieval pipeline design, chunking strategies, grounding outputs
• Anthropic model family fluency — Haiku, Sonnet, Opus tradeoffs across cost, speed, and quality
• AI-assisted development — using Claude Code to ship production-quality code; comfortable building tools that start as a conversation
• Multi-agent orchestration — chaining models, skills, and tools into coherent end-to-end workflows

Development

• Python — primary language for AI/ML work and automation
• JavaScript / Node.js for tooling and integrations
• FastAPI or Flask for internal service endpoints
• REST API design and consumption; webhooks and event-driven patterns
• Git and version control discipline
• Lightweight frontend — HTML/CSS, basic React, or comparable — for staff-facing tool interfaces

Data & Storage

• SQL — querying, joins, aggregations across relational databases
• PostgreSQL or MySQL in production environments
• Vector databases — Pinecone, Weaviate, pgvector, or equivalent
• JSON and structured data handling, transformation, and normalization
• Google Sheets and Excel, including scripting and automation

Deployment & Operations

• Lightweight deployment platforms — Vercel, Railway, Render, Fly.io, or similar (most internal tools live here)
• Cloud basics on AWS, GCP, or Azure for heavier workloads
• Docker and containerization
• CI/CD pipeline setup and maintenance
• Logging, alerting, and monitoring (Sentry, Datadog, LangSmith, or equivalent)
• Auth, multi-tenant access patterns, and basic security hygiene for internal tools

Qualifications

Required

• 3+ years building production LLM applications — not just proofs of concept
• Deep, hands-on prompt engineering experience across complex multi-step workflows
• Demonstrated ability to own a system end-to-end: build, ship, monitor, improve
• Strong judgment on AI output quality — you can identify when a model is wrong and you know how to fix it
• Experience integrating and maintaining third-party APIs in live environments
• Fluent written English and the ability to communicate clearly with non-technical stakeholders, translating ambiguous requests into working systems
• Strong async, autonomous work habits — you don't need a meeting to unblock yourself

Preferred

• Anthropic Claude Certified Architect (CCA) — strong signal but not required; the cert is new and

not yet widely held

• Experience evaluating and red-teaming AI systems for accuracy and safety
• Hands-on experience with multi-agent frameworks and autonomous workflow patterns
• Experience in a high-growth startup or small team where the engineer ships, not just specs
• Background in a domain where data accuracy has real consequences (health, finance, legal, regulated industries)
• Track record building tools specifically for internal non-technical users

Who Thrives in This Role This role rewards a specific kind of person. Read this carefully — if it doesn't sound like you, this probably isn't the right seat.

• You are obsessed with frontier AI. When Anthropic ships something new on a Tuesday, you've already played with it by Wednesday. You read changelogs. You follow the people doing the most interesting work. You don't need anyone to tell you what to learn next.
• You are deeply resourceful. You come from an environment where things had to work without a big team or big budget. You figure things out. You're allergic to "that's not my job."
• You ship fast and self-correct. You'd rather have something working in front of a stakeholder by end of day than a polished plan by end of week. When something breaks, you fix it without drama.
• You operate well in ambiguity. Specs from non-technical SMEs are sometimes incomplete. That's the job. You ask sharp questions, fill in the gaps, and ship.
• You treat internal tools like products. The fact that the user is on your team is not an excuse for things to be janky. Your tools work, look reasonable, and don't surprise people.
• You are honest about model failure. When an output is unreliable, you say so. You don't paper over it. You build the evals and guardrails to catch it next time.
• You write well and communicate async. Most of our team works across time zones. Clear writing is how you operate.

Logistics

• Reports to: CEO
• Location: Remote. International candidates encouraged. Some overlap with US Eastern hours expected, 8am to 12pm Eastern.
• Compensation: Senior engineer band, calibrated to candidate location and experience.
• Interview process: Initial conversation, technical deep-dive on a real Triquetra-style problem, take-home build, final conversations with stakeholders.

Apply To This Job