LLM Algorithm and Software Engine

About the Role We are a fast-growing semiconductor and AI startup developing high-performance Transformer ASICs for large language model (LLM) inference. We are looking for a talented Graduate AI Algorithm & Software Engineer who is passionate about LLMs, AI agents, and hardware/software co-design. This is a cross-disciplinary role sitting between LLM algorithms, AI software, applications, and hardware architecture. You will work closely with our chip architects and software engineers to understand the rapidly evolving LLM ecosystem, translate new model requirements into hardware and software requirements, and build agent-based applications running on our AI computing platform. This role is particularly suitable for a strong graduate who wants to understand AI systems from Transformer algorithms and Python models all the way down to AI accelerator hardware. What You Will Do

1. LLM Algorithm & Hardware Architecture Support

• Study the latest Transformer-based LLM architectures and inference techniques.
• Read, understand, modify, and prototype LLM implementations in Python and PyTorch.
• Analyse emerging models and identify their computational, memory, and communication requirements.
• Translate LLM workloads into requirements that can be understood and used by our AI accelerator and ASIC architecture teams.
• Analyse key Transformer operations including attention, MLP, MoE, KV cache, quantisation and matrix multiplication.
• Evaluate model characteristics such as parameter count, context length, memory footprint, bandwidth requirements, compute requirements and inference latency.
• Build Python-based models, benchmarks and tools to help evaluate architectural decisions.
• Keep the hardware team informed about important developments in the rapidly changing LLM ecosystem.

2. AI Agent Development

• Develop AI agents and agentic applications using modern open-source frameworks and libraries, including Pi Agent and similar agent frameworks.
• Build tool-using agents capable of interacting with APIs, software tools, databases and external services.
• Develop multi-step agent workflows for real-world applications.
• Experiment with techniques including tool calling, context management, memory, planning and multi-agent systems.
• Evaluate new open-source agent frameworks and identify technologies suitable for production deployment.

3. LLM Hardware/Software Solution Development

• Develop end-to-end hardware/software solutions for deploying LLM applications on our AI accelerator platform.
• Work with hardware, compiler, runtime and application engineers to integrate LLM models with our ASIC.
• Develop model deployment pipelines, runtime components, APIs, benchmarking tools and demonstration applications.
• Optimise LLM inference for latency, throughput, memory utilisation and energy efficiency.
• Help build reference applications demonstrating how customers can deploy LLMs and AI agents using our hardware.
• Support performance profiling, debugging and hardware/software co-optimisation. What We Are Looking For Essential
• Bachelor's or Master's degree in Computer Science, Artificial Intelligence, Electronic Engineering, Computer Engineering, Mathematics, or a related discipline.
• Strong programming skills in Python.
• Good understanding of Transformer architecture and large language models.
• Experience with PyTorch or another modern machine-learning framework.
• Ability to understand and modify open-source LLM code rather than treating models purely as black boxes.
• Good understanding of fundamental ML concepts, including attention, embeddings, matrix operations and neural-network inference.
• Strong problem-solving ability and willingness to work across traditional boundaries between algorithms, software and hardware.
• Curiosity about new LLM architectures and the ability to learn quickly in a rapidly evolving field. Desirable Experience in one or more of the following would be advantageous:
• Hugging Face Transformers
• Llama, Qwen, DeepSeek, Mistral or other open-source LLMs
• Mixture-of-Experts (MoE)
• KV-cache optimisation
• Quantisation, including INT8/INT4
• vLLM, SGLang or similar LLM inference frameworks
• AI agents and agentic workflows
• Pi Agent or other open-source agent frameworks
• CUDA, GPU programming or AI accelerators
• C/C++
• Linux development
• Docker and containerised deployment
• REST APIs and backend development
• Performance profiling and benchmarking
• Computer architecture, digital hardware or FPGA/ASIC development You do not need to have experience in all of these areas. We are looking for candidates with strong fundamentals, excellent learning ability, and genuine interest in working across the full AI computing stack. What Makes This Role Different Most graduate AI roles focus either on model development or application software. This position provides exposure to the entire LLM computing stack: LLM Models → Algorithms → Agents → Applications → Runtime/Software → AI Architecture → Transformer ASIC You will work directly with experienced AI and semiconductor engineers and have the opportunity to influence how future LLM workloads are implemented in silicon. As LLM architectures evolve, you will help answer questions such as:
• What will the next generation of LLMs require from AI hardware?
• Which operations should be accelerated directly in silicon?
• How should memory architecture evolve for long-context inference?
• How should MoE, attention and KV cache be handled efficiently?
• How can agentic AI applications take advantage of specialised inference hardware? Who This Role Is For This role may be a strong fit if you are a recent graduate who:
• enjoys reading and experimenting with new AI research and open-source models;
• can move comfortably between mathematical concepts and Python implementation;
• wants to understand how LLMs actually execute on hardware;
• enjoys building real applications rather than only training models;
• is interested in both AI agents and AI chips; and
• wants to work in an early-stage technology company where engineers can have significant technical ownership. Why Join Us You will join a team developing a new generation of Transformer inference ASICs designed specifically for large-scale LLM workloads. You will have the opportunity to work across AI algorithms, agentic systems, software and semiconductor architecture—gaining experience that is difficult to obtain in a conventional graduate software or machine-learning role. If you want to understand not only how to use LLMs, but how the computing systems underneath them should be designed, we would like to hear from you.