Intern - Grounded GenAI & Multimodal Knowledge Systems

James Hardie•Fontana, CA

7d•$22 - $25

About The Position

James Hardie is the industry leader in exterior home and outdoor living solutions, with a portfolio that includes fiber cement, fiber gypsum, composite and PVC decking and railing products. Our family of trusted brands includes Hardie®, TimberTech®, AZEK® Exteriors, Versatex®, fermacell®, and StruXure®. With over 8,000 employees and our U.S. operating entities headquartered in Chicago, we boast 31 operating sites, 6 recycling facilities, and 6 research and development centers globally. Powered by a dynamic workforce, we’re united by our purpose of Building a Better Future for All™ through sustainable innovation, a Zero Harm culture, and a commitment to empowering our people and communities. For more information, visit www.jameshardie.com . Summary As an intern at James Hardie, you will have the opportunity to gain hands-on experience in your chosen field, whether it be HR, Engineering, Finance, Marketing, IT, or another function. This is a 12-week paid internship program designed to provide you with valuable insights into the industry, enhance your skills, and prepare you for a successful career. You will work closely with experienced professionals, contributing to real projects that impact our business. Additionally, you will have access to exclusive mentoring opportunities to help guide your professional development. Internship Objective: This internship focuses on delivering a production‑ready prototype of a tightly grounded GenAI assembly‑guidance assistant while developing real‑world capability in RAG, knowledge structuring, evaluation, and safe, efficient system design using modern Azure‑based AI tooling. It directly advances departmental goals by improving customer safety and clarity, accelerating internal technical workflows, establishing scalable knowledge governance, and enabling cost‑disciplined deployment of AI‑powered support experiences.

Requirements

Must be enrolled in a bachelor's program in a relevant discipline at an accredited university, with an anticipated graduation date between December 2026 and May 2028 (rising junior or rising senior).
Demonstrated strong academic performance with a minimum GPA of 3.0.
Must be available to work full-time for 12 weeks during the summer of 2026.
Must have reliable transportation to and from the work location.
Must be authorized to work in the US.

Nice To Haves

Target major: Computer Science (AI/ML, NLP/Language Tech, Intelligent Systems), Data Science / Applied Data Science / Computational Data Science, Artificial Intelligence (where offered), Computational Linguistics (CS/Linguistics), Information Science / Informatics / Information Studies / Information Management
Familiarity with the following platform(s): Google Cloud, MS Azure, NVIDIA Nemotron, AWS Bedrock
RAG & Search: document ingestion/chunking, embeddings, hybrid retrieval (keyword + vector), reranking, and citation-grounded answer generation.
Document understanding: parsing PDFs/HTML/tables, text normalization, metadata extraction, deduplication, and version/applicability tagging for manuals/engineering docs.
Evaluation & verification: golden test set creation, automated regression tests, groundedness checks (citation coverage per step/claim), refusal/abstain gating, and failure mode analysis.
Multimodal evidence pipelines: video ASR/transcription, timestamped segmentation, indexing/search over segments, and tagging evidence objects to ontology entities with provenance (tier + review state).
Deployment & performance: Azure-based implementation, API integration with LLMs (Azure OpenAI + Gemini fallback), latency/cost profiling, caching/routing policies, and basic on-device inference workflows (quantization awareness/ONNX a plus).

Responsibilities

Work with team members on diverse projects and tasks.
Perform research and analysis to aid project objectives.
Create reports, presentations, and necessary documentation.
Engage in team meetings and share ideas.
Learn and implement industry best practices and company standards.
Build/Evaluate a controlled RAG pipeline over product manuals and engineering docs, including document ingestion, chunking, indexing, retrieval (hybrid where applicable), and reranking.
Implement safety/grounding controls: citations per step/claim, abstain/clarify behavior when evidence is insufficient, and a verification pass to detect missing evidence or conflicts.
Develop ontology + metadata schema for assembly-level reasoning (assemblies/subassemblies/parts/tools/specs/procedures/safety warnings) and apply it to improve retrieval precision and applicability filtering (model/variant/revision).
Create an evaluation harness (golden question set, rubrics, regression tests, dashboards) to measure groundedness, refusal correctness, safety, and answer usefulness over time.
Prototype a video-to-evidence workflow (internal mode): generate searchable evidence objects (transcripts + timestamps + segments + tags), and define a governance pathway to promote vetted knowledge into canonical sources.
Prototype an end-to-end research document ingestion pipeline for domain-expert reasoning of research literature, downstream experimental insights leveraging paper-qa, lab bench, nemotron-parse or similar tools.