Deskripsi Pekerjaan
Nexus Intelligence is building the cognitive layer for the modern enterprise. We are a well-funded, Series B startup backed by top-tier VCs (Sequoia, a16z) deploying LLMs at scale to automate complex knowledge work. We value velocity, craftsmanship, and intellectual honesty.
We are looking for a Senior AI/ML Engineer to own the lifecycle of our core foundation models and RAG pipelines. You will not just call APIs; you will fine-tune, optimize inference, and architect the data flywheels that power our agentic platform. This is a high-autonomy role for an engineer who wants to define the state-of-the-art in production AI.
Why This Role Is Different
- GPU-Rich Environment: Dedicated access to H100/A100 clusters; no ticket queues for compute.
- Open Source Contribution: Paid time to upstream improvements to vLLM, Hugging Face, LangChain, or PyTorch.
- End-to-End Ownership: From data curation and pre-training to low-latency serving and eval frameworks.
- Hybrid Flexibility: 3 days/week in our sunny SoMa loft (walking distance to BART/CalTrain), 2 days remote.
Tanggung Jawab
- Design, train, and optimize LLMs (7B–70B parameter range) for domain-specific reasoning and tool use.
- Architect and scale Retrieval-Augmented Generation (RAG) systems: chunking strategies, hybrid search, re-ranking, and citation verification.
- Build robust evaluation harnesses (unit tests for LLMs, human-in-the-loop annotation pipelines, automated regression detection).
- Optimize inference latency and throughput using quantization (GPTQ/AWQ), speculative decoding, and custom CUDA kernels via Triton.
- Own the MLOps pipeline: experiment tracking (Weights & Biases/MLflow), CI/CD for model deployment (Kubeflow/Argo), and feature stores.
- Collaborate with Product and Design to define AI-native UX patterns (streaming, structured output, agent orchestration).
- Mentor junior engineers and establish best practices for code review, testing, and documentation in an AI-first codebase.
- Stay at the bleeding edge: reproduce papers, benchmark new architectures (Mamba, RWKV, Mixture-of-Experts), and advise leadership on technical strategy.
Kualifikasi
- 5+ years of professional software engineering experience; 3+ years hands-on with Deep Learning in production.
- Expert-level Python; strong proficiency in PyTorch (or JAX/TensorFlow). Comfortable reading and writing C++/CUDA for performance hotspots.
- Proven track record of training/fine-tuning LLMs (SFT, DPO/RLHF, LoRA/QLoRA) on multi-node GPU clusters.
- Deep understanding of Transformer architecture, attention mechanics, positional encodings, and scaling laws.
- Production experience with vector databases (Pinecone, Weaviate, Milvus, PGVector) and semantic search optimization.
- Solid software engineering fundamentals: Docker, Kubernetes, Linux, Git, CI/CD, observability (Prometheus/Grafana/Datadog).
- Bachelor’s/Master’s/PhD in Computer Science, Machine Learning, or equivalent practical experience (open-source contributions, Kaggle Grandmaster, top-tier publications).
- Excellent communication skills; ability to explain complex trade-offs to non-technical stakeholders.