Artificial Intelligence Engineer

We are looking for an experienced AI / LLM Engineer to join and help build the intelligence layer behind our AI-powered GRC and cybersecurity platform.This is not a generic machine learning role. You will work hands-on with open-weight Large Language Models, fine-tuning and optimizing them for our domain and progressively evolving them into a highly specialized proprietary AI layer.The role spans the full LLM lifecycle: dataset engineering, fine-tuning, evaluation, inference optimization, RAG, tool use, reasoning workflows, model serving, and production deployment.You will work on turning strong foundation models into AI systems that understand Ossus-specific domains, terminology, workflows, policies, controls, evidence, regulations, and security context.Key ResponsibilitiesFine-tune and adapt open-weight LLMs for specific use casesBuild and maintain domain-specific training datasets from structured and unstructured dataDesign instruction-tuning, supervised fine-tuning, LoRA/QLoRA, preference-tuning, and other model adaptation pipelinesDevelop rigorous LLM evaluation frameworks to measure reasoning quality, factual accuracy, hallucination rate, instruction following, domain knowledge, and task performanceBenchmark different model versions, fine-tuning techniques, prompts, quantization methods, and inference configurationsOptimize models for quality, latency, throughput, GPU memory consumption, and costWork with inference technologies such as vLLM, Hugging Face Transformers, CUDA-based inference stacks, and optimized model runtimesExperiment with and deploy BF16, FP8, FP4 and other quantization / precision strategies where appropriateBuild sophisticated Retrieval-Augmented Generation (RAG) systems over regulatory documents, policies, controls, evidence, organizational data, and knowledge basesDesign document ingestion, chunking, metadata, embedding, retrieval, reranking, and context-selection pipelinesDevelop systems combining LLMs, RAG, tools, APIs, structured outputs, and multi-step reasoningBuild mechanisms for grounding model responses in authoritative sources and reducing hallucinationsDevelop AI workflows that can reason across multiple documents, regulations, controls, findings, and pieces of evidenceDesign and maintain prompt, context, memory, and orchestration layers around the underlying modelsHelp evolve individual models into a larger Ossus proprietary AI architecture rather than relying solely on third-party AI APIsBuild synthetic data generation and data augmentation pipelines when useful for model trainingCreate automated regression tests so model improvements do not degrade existing capabilitiesDevelop APIs and services exposing AI capabilities to the wider Ossus platformProfile and troubleshoot GPU inference, model-loading, memory, concurrency, and performance issuesCollaborate closely with backend, product, security, and GRC teams to translate complex domain requirements into reliable AI capabilitiesResearch new LLM techniques and determine which are genuinely useful enough to incorporate into productionWrite clean, maintainable, testable, and well-documented Python codeRequired Qualifications3+ years of professional experience in AI, Machine Learning, NLP, LLM engineering, or a closely related engineering roleStrong proficiency in PythonStrong practical understanding of Large Language Models and Transformer architecturesHands-on experience working with PyTorchExperience with Hugging Face Transformers and the Hugging Face ecosystemHands-on experience fine-tuning or adapting open-source / open-weight LLMsUnderstanding of techniques such as LoRA, QLoRA, PEFT, supervised fine-tuning, and instruction tuningStrong understanding of model training, validation, benchmarking, and evaluationExperience preparing and curating datasets for model trainingUnderstanding of tokenization, context windows, attention, embeddings, inference, and model precisionExperience deploying and serving models on GPUsExperience diagnosing model quality and inference-performance issuesExperience with Git, Docker, Linux, and modern software engineering practicesAbility to independently research technical approaches, reproduce relevant research, and determine whether an approach provides measurable improvementsStrong analytical and experimental mindsetPreferred QualificationsExperience with several of the following would be a strong advantage:Open-weight model familiesvLLMNVIDIA GPUs and CUDAMulti-GPU inference and trainingBF16, FP16, FP8, FP4, AWQ, GPTQ or other model quantization techniquesDistributed training and inferenceRAG architecturesEmbedding models and vector databasesHybrid retrieval and semantic searchReranking modelsAdvanced document parsing and ingestionLong-context LLM applicationsLLM agents and tool callingStructured output / constrained generationSynthetic dataset generationAutomated LLM evaluation and benchmarkingPreference optimization techniques such as DPO or related approachesLLM observability and production monitoringLangChain, LlamaIndex, DSPy, or similar frameworksREST APIs and backend service developmentGCP, AWS, or AzureMLOps and CI/CD for AI systemsSQL and database technologiesCybersecurity, GRC, compliance, or regulatory technologyWhat You’ll Be BuildingOur objective is not simply to connect Ossus to an external LLM API.We are building an increasingly specialized Ossus AI layer that combines:Foundation Models → Domain Fine-Tuning → Ossus Knowledge → RAG → Tools → Reasoning → Evaluation → Continuous ImprovementOver time, the underlying models should become increasingly capable of understanding and reasoning about areas such as:Cybersecurity frameworks and controlsRegulatory requirementsGovernance, Risk, and ComplianceOrganizational policies and proceduresSecurity evidence and technical findingsAudits and assessmentsControl mappings across different frameworksCompliance gaps and remediationLarge collections of enterprise and regulatory documentsYou will play a significant role in designing how that intelligence layer evolves.What We’re Looking ForWe are looking for someone who is both an AI researcher-minded engineer and a strong practical builder.You should be comfortable going from:research paper → experiment → benchmark → fine-tuned model → optimized inference → production system.We care much more about measurable model improvements and working systems than buzzwords or simply connecting APIs together.You should enjoy experimenting with models, questioning whether an approach actually improves results, digging into why models fail, optimizing GPU performance, building strong evaluation datasets, and progressively making the system smarter.We value technical ownership, curiosity, experimentation, rigorous evaluation, clean engineering, and practical problem-solving.