Senior Python AI Engineer, LLM / Agents

At Variant Group, we build B2C SaaS products that millions of people use every day — from résumé builders to legal document generators. We turn complex challenges into simple, accessible solutions.

Variant Group Teams:

• pdf.net — 30 members

• resume.co, resumetemplates.net & cv.co — 15 members

• mealplan.co — 5 members

Job

pdf.net is the fastest-growing product in our portfolio. Thanks to bold product breakthroughs, it already serves hundreds of thousands of users every month, and is on track for even greater growth.

We’re looking for a Senior Applied AI Engineer to build production-grade AI systems for document workflows at a massive scale. You will design and ship agentic, multimodal, evaluation-driven product features across extraction, reasoning, generation, and orchestration — owning everything from architecture and experimentation to latency, cost, reliability, and user impact.

This is a hybrid role that blends custom model development with LLM API integration to ship intelligent, production-ready features. You’ll work across the full lifecycle — from preparing training data and fine-tuning models, to designing retrieval pipelines and deploying performant inference systems in the cloud.

What You’ll Do

• Work on document intelligence problems: OCR, structured extraction, chunking, layout-aware parsing, and PDF-to-schema workflows.

• Design and ship RAG pipelines using vector databases, hybrid retrieval, reranking, and structured context assembly.

• Integrate hosted LLM APIs (OpenAI, Anthropic) and, where needed, smaller self-hosted models into product features.

• Build and fine-tune transformer models for classification, extraction, ranking, and generation tasks.

• Build scalable inference services using vLLM, batching, streaming, caching, and other latency/cost optimizations.

• Implement production pipelines for training, evaluation, prompt/version management, and deployment.

• Define and improve AI quality with offline evals, regression datasets, and task-level metrics for accuracy, latency, and reliability.

• Own AI-related infrastructure in Docker, Cloud Run, and GCP, with proper logging, monitoring, and observability.

• Collaborate with product, design, and full-stack engineers to ship user-facing AI features end to end.

What You Bring

• Strong Python engineering background with production-grade code, tests, and maintainable service architecture.

• Hands-on experience with PyTorch and HuggingFace, including training, fine-tuning, inference, and model debugging.

• Solid understanding of transformer internals, tokenization, embeddings, decoding strategies, and evaluation trade-offs.

• Experience integrating OpenAI / Anthropic APIs into real user-facing products, not just prototypes.

• Practical experience with RAG systems: chunking, embedding pipelines, vector search, reranking, context construction, and hallucination reduction.

• Experience with vector stores such as pgvector or FAISS; understanding of when retrieval helps and when deterministic pipelines are better.

• Familiarity with document understanding workflows: OCR, PDF parsing, table/form extraction, and normalization of messy real-world inputs.

• Experience building scalable inference or model-serving systems; familiarity with vLLM or similar serving stacks is a plus.

• Ability to debug model quality issues across prompts, retrieval, data, evaluation sets, and inference behavior.

• Comfortable owning systems in production: deployment, monitoring, incident debugging, and iterative improvement.

Bonus Points For

• Experience deploying ML/AI services with Docker on GCP (Cloud Run, GKE).

• Experience with OpenTelemetry, Prometheus, Grafana, or similar observability tooling.

• Experience with Neo4j or other graph-based knowledge systems where graph retrieval actually improves product behavior.

• Experience with RLHF, preference data, or ranking/reward modeling.

• Strong background in document understanding, OCR, and structured PDF parsing.

• Experience optimizing inference for latency and cost using batching, caching, routing, or speculative decoding.

Working with us

Pure Data-Driven SaaS

Learn and experiment, no opinion-driven decisions.

Location Independence

Work from a location that suits you - be it your home, a café, or anywhere else.

Unlimited Time Off

We trust you to take time off when you need, ensuring you’re always at your best when you’re with us.

Results-Oriented

We believe in the quality of output over hours clocked in. Deliver excellence, and enjoy unparalleled autonomy in return.

Salary & Contract

• Compensation: $60,000 — $132,000/year gross.

• Paid via deel.com (standard contract).

• Asia/Pacific or EU timezone.

How to apply?

1. Submit via Google Forms (5-8 mins).

2. After a successful resume review, you'll be invited for an interview. If needed, we may ask you to complete a short test assignment (2–4 hours maximum).