What You’ll Do

• Integrate hosted LLM APIs (e.g. OpenAI, Anthropic) + custom models to support intelligent in-product behavior.

• Build and fine-tune transformer models using PyTorch, HuggingFace

• Design and deploy retrieval-augmented generation (RAG) pipelines with vector databases (e.g., pgvector) and graph-based reasoning (e.g., Neo4j).

• Develop scalable inference systems using vLLM, speculative decoding, and optimized serving techniques.

• Build modular, production-grade pipelines for training, evaluation, and deployment.

• Collaborate closely with product, design, and full-stack teams to ship features that bring AI to end users.

• Own infrastructure around Docker, Cloud Run, and GCP, ensuring speed, reliability, and observability.

What You Bring

• Strong Python engineering background with clean, tested, and maintainable code.

• Proven experience building with transformer-based models, including custom training and fine-tuning.

• Deep familiarity with HuggingFace, PyTorch, tokenization, and evaluation frameworks.

• Experience integrating and orchestrating LLM APIs (OpenAI, Anthropic) into user-facing products.

• Understanding of semantic search, vector storage (FAISS, pgvector), and hybrid symbolic-neural approaches.

• Experience designing or consuming graph-based knowledge systems (e.g., Neo4j, property graphs).

• Ability to build and debug scalable training and inference systems.

Bonus Points For

• Hands-on experience with Docker and production deployment on Google Cloud (GKE, Cloud Run).

• Experience with RLHF, reward models, or reinforcement learning for LLM alignment.

• Knowledge of document understanding, OCR, or structured PDF parsing.

• Exposure to monitoring and observability tools (e.g., Prometheus, Grafana, OpenTelemetry).

• Background in linguistics, semantics, or computational reasoning.