Deep Learning Engineer

About Us:

Trail is the context layer AI agents run on. We're building the missing layer between your company's rules, exceptions, and institutional knowledge and the agents trying to act on them. It's a governed graph that lets any agent or platform, including Trail agents, make the right call on your most complex business processes. Agents today fail not because models are weak, but because they're never given the judgment a business builds up over years. We're fixing that.

Visit our Website and Youtube to learn more!

About the Role

The role can be summed up as building and deploying cutting edge generalised deep learning architectures that can solve complex business problems like converting unstructured data into structured format without hand-tuning features/models. You are expected to build state of the art models that are best in the world for solving these problems, continuously experimenting and incorporating new advancements in the field into these architectures.

What we’re looking for

• 5-8 years of experience in Deep Learning.

• Strong foundational knowledge in deep learning concepts and architectures (LLMs and VLMs)

• Demonstrated expertise in at least one specialised area of deep learning (NLP, computer vision, multimodal models, etc.)

• Experience building and deploying production-grade Deep Learning systems at scale,

• Familiarity with various large language models (GPT, LLaMA, Claude, etc.) and their applications

• Strong software engineering practices including version control, CI/CD, and code quality

• Ability to rapidly learn and apply new technologies and approaches.

Interesting Projects Other Senior DL Engineers Have Completed

• Deployed large scale multi-modal architectures that can understand both text and images really well.

• Built an auto-ML platform that can automatically select the best architecture, fine-tuning method based on type and amount of data.

• Best in the world models to process documents like invoices, receipts, passports, driving licenses, etc.

• Hierarchical information extraction from documents. Robust modeling for the tree-like structure of sections inside sections in documents.

• Extracting complex tables — wrapped around tables, multiple fields in a single column, cells spanning multiple columns, tables in warped images, etc.

• Enabling few-shots learning by SOTA finetuning techniques.