LLM Research Intern

About ProCogia: 

ProCogia is a data and AI consulting firm helping organizations turn complex technology challenges into measurable business outcomes. We work across highly regulated and high-stakes industries, including telecommunications, financial services, life sciences, healthcare, and the public sector, where security, governance, reliability, and performance are critical. Our teams bring together expertise across data science, engineering, AI, analytics, and consulting to design and deliver practical, production-ready solutions. We are actively advancing capabilities across generative AI, agentic systems, workflow automation, AI governance, and enterprise AI applications. At ProCogia, you will work alongside experienced practitioners on meaningful technical challenges and contribute directly to both client solutions and the platforms and products we are building internally.

Our Core Values: Trust, Growth, Innovation, Excellence, and Ownership

Work Location: Vancouver (On-site – Full-Time Work from office in Vancouver, five days a week Mon-Fri)  

Job Summary 

ProCogia is looking for a curious and technically strong LLM Research Intern to join our Vancouver team. This is a hands-on research and engineering role for someone interested in understanding how large language models can be adapted, evaluated, optimized, and deployed for specialized enterprise use cases. You will work alongside our AI, data science, and engineering teams to experiment with open-weight LLMs, supervised fine-tuning, continued pretraining, parameter-efficient adaptation, model evaluation, RAG, and inference optimization.

Key Responsibilities 

• Assess domain-specific datasets and determine the right adaptation approach across fine-tuning, continued pretraining, RAG, or hybrid strategies.
• Establish baseline performance and measure whether model adaptations produce meaningful, defensible improvements.
• Curate, clean, deduplicate, structure, and quality-filter datasets for training and evaluation.
• Fine-tune open-weight LLMs using techniques such as LoRA, QLoRA, PEFT, and multi-adapter approaches.
• Support training and experimentation across single-GPU and distributed multi-GPU environments.
• Design rigorous evaluation frameworks covering factuality, reasoning, grounding, robustness, domain performance, and failure modes.
• Apply public benchmarks and build task-specific evaluation datasets and metrics where standard benchmarks are insufficient.
• Compare models, prompting strategies, retrieval methods, and adaptation techniques through controlled experiments.
• Analyze trade-offs across model quality, accuracy, latency, memory, GPU utilization, token usage, compute requirements, and cost.
• Improve model reliability by evaluating RAG, grounding, reranking, guardrails, hallucination-reduction techniques, and inference optimization.