Note: The job is a remote job and is open to candidates in USA. The C++ Alliance, Inc. is a nonprofit foundation supporting the C++ programming language and its open-source ecosystem. The organization is hiring a senior Python and AI Pipeline Engineer to build production AI infrastructure, including document-processing pipelines, classifiers, self-hosted LLM systems, editorial workflows, and legacy system modernization. The role emphasizes reliable, deterministic, observable, and reproducible engineering for technical content analysis.
Responsibilities
- Document processing pipelines. Conversion of papers and other technical content from PDF, HTML, and LaTeX into structured markdown suitable for downstream analysis. Quality matters — these conversions feed everything else, so the pipelines are built with human approval gates and validation loops
- Classifier engineering for technical content. Multi-label sentence and document classifiers for analyzing standards committee papers — categorizing paper types, identifying structural features (rationale, prior art citations, implementation evidence), and surfacing analytical signal at scale. The work involves golden set preparation, fine-tuning embedding-based classifiers, evaluation harnesses, and the operational infrastructure to run inference reliably and deterministically
- Self-hosted LLM infrastructure. Deployment of open-weight models (Qwen, Gemma, and others) on our own cloud infrastructure. This is non-negotiable for the project — production analysis needs to be deterministic and reproducible, which means we cannot use shared cloud LLM endpoints where service quality can vary invisibly between requests. The work includes model serving, fine-tuning workflows, evaluation infrastructure, and making self-hosted models actually usable in production
- Editorial workflow systems. Pipelines that integrate AI-generated analysis with human review on community mailing lists, including content ingestion, multi-stage processing with human approval gates, and editorial selection workflows. The work involves agent orchestration where appropriate, but primarily focuses on systems that augment human reviewers rather than replace them
- Legacy system modernization. The community infrastructure we are building on includes long-running open-source projects (Mailman 3, Postorius, and others) that need authentication integration, feature enhancement, and the kind of careful upstream-compatible modification that does not destabilize a 40-year-old codebase
Skills
- Strong candidates may emphasize one or more of: Classical ML and classifier engineering. You have built and shipped production ML systems — fine-tuned classifiers, embedding-based retrieval, multi-label sentence-level analysis. You know how to prepare golden sets, evaluate model performance honestly, and build deterministic inference pipelines that produce stable results. BERT-family fine-tuning, scikit-learn, PyTorch, and the rest are tools you have used in production, not just in coursework
- Production LLM application engineering. You have shipped real AI features to real users — RAG pipelines, structured generation, evaluation frameworks, prompt engineering. You know the difference between a working demo and a system that holds up in production, and you understand when frontier models are the right tool and when they are not
- LLM infrastructure and serving. You have deployed models — vLLM, Triton, multi-model routing, inference optimization, evaluation pipelines. You think about cost-per-token, latency tails, determinism, and how to keep complex serving systems reliable
- Production pipeline engineering. You build multi-stage workflows with proper error handling, observability, and operational discipline. ETL is in your background, even if the new pipelines have ML components in them. You think about what happens when stage three fails after stage two has committed, and about human-in-the-loop gates that prevent bad output from reaching users
- Real production experience. Demos and prototypes are not the same as systems that have served real users for months. We are looking for candidates who have shipped, maintained, modernized, and lived with the consequences
- Honesty about what you have done. AI engineering is full of overclaim right now. We have a strong preference for candidates who describe their actual contributions accurately, including what worked and what did not
- Engineering rigor. Tests, documentation, monitoring, observability, careful failure-mode thinking. ML and AI systems fail in subtle ways and we want engineers who care about catching those failures before they reach users
- Determinism and reproducibility. Our domain is one where non-deterministic outputs damage trust. We prefer engineers who think carefully about reproducibility, who choose deterministic tools when possible, and who know how to make probabilistic systems behave as predictably as the use case requires
- Comfort with the open-source ecosystem. The Alliance exists to support open-source C++ work. Some familiarity with — or curiosity about — the open-source world helps
Benefits
- Remote work
- Preference for candidates in Pacific time or with substantial overlap into Pacific working hours
Company Overview