LLM Application Engineer (Remote, Monthly Retainer)

We're hiring a mid-senior LLM Application Engineer on a remote monthly retainer to design, ship, and harden production AI pipelines for client products at Bolder Apps. You'll own structured extraction and classification systems that turn messy real-world inputs (email, HTML, PDFs, images) into reliable product data, with measurable quality gates, evaluations, and cost control.

You'll work on Firebase / GCP-style backends with product, mobile, and QA teams. We want someone who has shipped LLM apps for real users, not just demos. You should be strong across modern LLMs and especially fluent with Google Gemini (multimodal prompts, structured outputs, failure modes, and cost/latency tradeoffs), with solid experience on other major providers too. If you can hit hard quality targets, keep dollars-per-run honest, and leave runbooks another engineer can pick up, we want to talk.

Responsibilities

Key Responsibilities:

• Own production LLM pipelines end-to-end: ingestion, multimodal model calls, structured records, and storage, including confidence flags, retries, and idempotent rescans
• Design prompt and schema strategies (including schema-aligned or constrained outputs) to ensure consistent, product-ready results
• Build classification and filtering layers on top of extraction (taxonomy mapping, demographic/audience filters, deduplication, and related cleanup logic)
• Define and run evaluation harnesses (golden sets, regression suites, online metrics) to prevent quality regression when prompts, models, or parsers change
• Hit and report against hard quality targets (precision-style gates for completeness, duplicates, incorrect inclusions, image presence, and similar product SLAs)
• Optimize token usage, model tiering, caching, and batching to control dollars-per-run and latency
• Harden reliability for long-running async jobs (timeouts, partial recovery, memory limits, safe production deploys)
• Partner with Flutter/mobile and QA teams on field contracts, review queues, and incident debugging
• Document architecture and runbooks to ensure shared ownership, not single-point failure
• Stay current on Gemini and peer LLM APIs; recommend model swaps, fallbacks (e.g., document AI), or schema tightening

Required Experience:

• Shipped LLM applications in production (not demos only): prompts, structured outputs, retries, observability, and real failure handling
• Strong hands-on experience with Google Gemini, including multimodal (text + image/document-style) workflows and structured extraction
• Practical experience with at least one other major LLM stack (OpenAI, Anthropic, or similar) and good judgment on when to use which
• Structured extraction from messy inputs: HTML, PDFs, images, and mixed email-like content
• Classification/taxonomy systems on top of LLM outputs
• Evaluation discipline: offline evals, regression suites, and production quality metrics tied to clear acceptance criteria
• Cost and latency awareness: token budgeting, cheaper tiers, caching, batching; can explain dollars-per-run tradeoffs to a PM
• Python backend experience on serverless cloud (Cloud Functions or equivalent) and document stores (e.g., Firestore) or similar GCP patterns
• English at C1 or above for client-adjacent debugging with a PM
• US hours overlap through roughly 5 PM EST for live coordination
• Ownership habits: honest estimates, early blockers, finished releases

Nice to Have:

• Schema-aligned LLM frameworks (BAML, Instructor, Outlines, or similar)
• Google Document AI or other OCR/document intelligence as a fallback path
• Gmail API/OAuth products and restricted-scope compliance familiarity
• Computer vision for product-image quality checks
• Firebase + GCP operations (secrets, regions, schedulers, cost monitoring)
• Building eval corpora from real production data and iterating until contractual SLAs pass
• Agency or multi-client studio experience

Benefits

• Fully remote and async-friendly, with required overlap through ~5 PM EST for client/release coordination
• Monthly retainer structure for recurring AI pipeline work, emphasizing production quality and cost honesty
• Real autonomy over prompts, schemas, evals, and deploys—no rigid playbook
• Direct line to PMs, mobile engineers, and decision-makers
• Tooling budget for LLM and cloud tools to accelerate work
• Peer network of product-minded builders across overlapping client projects