Senior AI Research Engineer (Foundation… at Ailytics · Singapore

Here at Ailytics, we're building AI solutions to envision a safer world. By combining computer vision and predictive analytics, we enable organizations to proactively identify risks, optimize processes, and ultimately save lives. Our platforms are currently deployed all over the world, covering more than 500 million square meters!

We already have real-time generative AI in production: capabilities that let our customers search intelligently across their video, and define and run custom use cases and pipelines for their own environments. What we need now is someone to own it, scale it, and decide where it goes next.

This is the role for an engineer who wants the whole problem rather than a slice of it. You will fine-tune and adapt LLMs, VLMs and large vision models to our domain, build new functionality on top of them, and lay down the architecture everything else is built on. You will also own deployment: these models reach production through you, not over a handoff wall.

We are deliberately giving this role full flexibility to explore. You decide which approaches are worth pursuing, which models to build on, and what the architecture should look like, working directly with our CTO, Product Managers and senior management. We care about outcomes in the product, not about how much of your time looked like research versus engineering.

The one thing we are firm about is mindset. This is a startup, and we need someone product-driven: rigorous enough to know what will actually hold up in production, and pragmatic enough to ship it. You will work directly with our CTO, Product Managers and senior management, and what you build will shape the company’s direction. As the capability grows, you will set up and lead the team underneath you — hiring the engineers who scale it with you.

One more thing worth knowing: you will be working with a proprietary, large-scale, real-world video dataset spanning industrial environments across the world. It is the kind of data you cannot get from a public benchmark, and it is a large part of what makes the problems here interesting.

Location: This role is open to candidates based in either Singapore or Hong Kong. We are hiring for one position and are flexible on location across both markets.

What You Will Be Doing

• Fine-tune, adapt and experiment with LLMs, VLMs, large vision models and other foundation models to meet specific product and accuracy requirements.
• Build new functionality. Design and deliver genuinely new generative AI features, from first idea to something customers use.
• Optimise what exists. Push accuracy, robustness, latency and inference cost on our deployed models. These run in real time, so latency and throughput are first-class constraints rather than afterthoughts.
• Own the architecture and the deployment. Define how our generative AI systems are structured and integrate with our computer vision stack, and take models from experiment to production.
• Build the evaluation discipline. Design rigorous benchmarks and eval harnesses so we know what is actually improving, and can prove it.
• Curate and refine datasets: training and evaluation data, augmentation strategies, labelling quality and edge-case handling.
• Set the standard and build the team. Establish how generative AI work is done here, then hire and mentor the engineers who scale the function with you.

Our Ideal Requirements

The fundamentals:

• 5+ years as a computer vision or machine learning engineer, with at least 2 years focused specifically on generative AI.
• Strong proficiency in Python, with solid grounding in data structures, algorithms and software engineering principles. You write code others can build on.
• Hands-on, current experience fine-tuning open-weight foundation models such as VLMs (Qwen-VL, InternVL, LLaVA), LLMs (Qwen, Llama, Mistral, DeepSeek) or vision foundation models (DINOv2, SAM, Grounding DINO). What matters is that you have done this work yourself and can talk through it in depth, including what did not work.

Inference and serving depth:

• Deep understanding of LLM inference internals. Attention, KV-cache mechanics and decoding strategies. You understand why a model behaves the way it does at inference time, not just how to call it.
• Command of token and context management: tokenisation, context-window extension and its trade-offs, long-context behaviour, and budgeting context length against latency, throughput and cost.
• Experience serving models in production, including multi-GPU sharding, memory and throughput optimisation (vLLM, SGLang, TensorRT-LLM or similar), and the trade-offs between accuracy, latency, throughput and cost. You can diagnose an OOM or a throughput bottleneck rather than reaching for bigger hardware.
• Experience with edge or on-premise inference is a strong advantage. Running models on constrained hardware (NVIDIA Jetson or similar), quantising for deployment, and getting real-time performance out of limited compute. A significant part of our stack runs on customer premises rather than in a datacenter.

Applied generative AI:

• You have built or shipped a scalable LLM-based application. Something real users depended on, not a notebook or a demo.
• Deep familiarity with PyTorch and the Hugging Face ecosystem, and practical command of LoRA/QLoRA, supervised fine-tuning, quantisation and distillation.
• A good understanding of MCP and agentic AI: tool use, orchestration, multi-step agent design, and clear judgement about where agentic approaches genuinely help versus where they just add fragility.
• Strong evaluation and experiment design. You know how to build an eval that tells you the truth rather than one that flatters the model.
• Experience with multimodal systems (vision and language), which is central to what we do here.

Experience and ways of working:

• Bachelor's, Master's or PhD in Computer Science, Electrical Engineering or a related technical field.
• Experience architecting systems, not just models. Comfortable owning technical design decisions and defending them.
• Appetite to build a team. Experience mentoring or leading engineers, and a genuine desire to grow a function rather than remain a pure individual contributor.

Why Work At Ailytics?

• Lasting impact at the intersection of AI and safety. Your models are used daily by enterprise customers across Asia and beyond, in environments where a missed detection genuinely matters.
• We're a startup still in its early stages, so there's real room to shape both the product and the way we build it.
• You own generative AI here. Not a workstream within someone else's roadmap, but the capability itself, including the architecture and the direction.
• Support to stay at the frontier. A learning budget, and backing to attend and publish at the conferences that matter in this field.
• Keeping current is part of the job here, not something you do in your own time.

We value strong standards, high transparency, and low egos. Let us know if this sounds like you.