Senior NLP Engineer – Large Language Models Specialist – Remote Undefined

This is a remote position

What we do :

We are at the forefront of leveraging NLP technologies to revolutionize data analysis and decision-making

processes across various domains, including pharmaceuticals and research. Our dynamic team is currently

transitioning to advanced platforms such as Sinequa on Amazon SageMaker, enhancing our capabilities in handling

large-scale data with sophisticated language models.

Job Description :

We seek an experienced Senior NLP Engineer / Large Language Model Architect to lead the development and

optimization of our NLP projects. The ideal candidate will be instrumental in refining our large language models,

particularly focusing on areas such as pharmaceutical SOPs and research contexts. This role requires a deep

understanding of NLP principles, practical experience in deploying large-scale models, and a problem-solving

mindset geared towards delivering high-quality, efficient solutions.

Key Responsibilities :

Design, develop, and refine large language models to meet specific project requirements, ensuring high

quality and efficiency.

Lead the transition and optimization of NLP models from platforms like Solr and Elasticsearch to Sinequa

and Amazon SageMaker.

Develop and implement strategies for prompt engineering, model refinement, and training pipelines to

enhance model performance.

Collaborate with internal teams and stakeholders to understand project needs, providing expert guidance

and NLP architectural advice.

Manage the integration of knowledge graphs into NLP models to improve contextual understanding and

output relevance.

Evaluate and utilize state-of-the-art embedding vectors and encoding methods to ensure optimal model

performance.

Oversee the concept extraction and entity recognition processes, ensuring integration with existing

knowledge graphs and data pipelines.

Guide the team in the expansion and refinement of taxonomies using large language models, followed by

human review for tagging accuracy.

Drive the adoption of best practices in NLP model development, deployment, and maintenance, staying

abreast of the latest industry trends and research.

Required Skills and Experience :

Advanced degree in Computer Science, Artificial Intelligence, Linguistics, or a related field.

Extensive experience in NLP, with a strong portfolio of projects involving large language models.

Proficiency in using platforms such as Hugging Face, Sinequa, and Amazon SageMaker for model

deployment and refinement.

Solid understanding of model training, tuning, and refinement techniques, with proven success in applying

these to large-scale projects.

Experience with knowledge graphs, entity recognition, and concept extraction technologies.

Strong coding skills in relevant programming languages (e.g., Python) and familiarity with NLP libraries and

frameworks.

Demonstrated ability to work effectively in a fast-paced, dynamic environment, managing multiple

projects simultaneously.

Excellent problem-solving skills, with a keen eye for detail and a commitment to high-quality outcomes.

Strong communication and collaboration skills, with the ability to convey complex technical concepts to

non-technical stakeholders.

Desirable Attributes :

Experience in the pharmaceutical or research sectors, with an understanding of SOPs and compliance

requirements.

A pragmatic approach to project management, avoiding magical solutions and focusing on tangible

results.

The flexibility to adapt to changing project needs and the ability to guide teams through technical

challenges.

A team player attitude, with experience working in or leading multidisciplinary teams.

About the project & the Team :

We have a RAG platform. Currently, we are transitioning from Solr and Elasticsearch on Cloudera to Sinequa and

SageMaker on Amazon, not by choice, but it is what it is. We require NLP experts to handle aspects related to

large language models. Deploying the models is relatively straightforward, typically done through Hugging Face,

but that’s not the main issue. The challenge lies in the nuances of developing high-quality large language models,

including organizing prompts, refining training pipelines, and understanding when they work or fail. For instance,

over the last three weeks, we attempted refinement training on pharmaceutical standard operating procedures

SOPs), which was unsuccessful. Initially, the model forgot its chatbot functionalities. After two weeks of retraining

on a local cluster, the refinements on SOPs were lost again, indicating a fundamental flaw in our approach. Our

NLP team suggests experimental methods based on academic papers, but we lack the luxury of time for such trials.

Therefore, we have the need for experienced individuals who can provide architect-level guidance on managing

multiple large language models, including those specific to SOPs and research contexts. We are developing own

large language model using our data, which we plan to utilize. We need to strategize on prompt engineering and

ETL processes, and how to integrate our knowledge graph into the prompts. Currently, we are using the top

embedding vector from Hugging Face for encoding, which is feasible as our beta environment only hosts 20 million

records, allowing for a manageable re-encoding timeline.

We aim for high-quality outcomes with our large language models, applicable universally, not just our data. We

are seeking someone well-versed in the theoretical and practical aspects of NLP, capable of guiding us in

refinement training, estimating costs, and expected quality improvements. The goal is to have mature, pragmatic

discussions rather than rely on uncertain, magical solutions.

We are in a situation, recognizing that the field attracts many who prefer experimenting with technology over

delivering tangible value. We have learned that finding individuals with a comprehensive skill set is unrealistic due to

the multidisciplinary nature of this field. We have shifted our strategy to engage specialists for specific phases,

thereby avoiding the inefficiency of prolonged meetings.

Our requirement is straightforward : an expert who can effectively deploy large language models of high quality.

We are exploring various tools and methods, such as Data IKU for concept extraction and SciByte for named entity

recognition (NER), supported by our knowledge graph. Our challenges include managing taxonomies, expanding

synonyms with large language models, and ensuring human review for tagging accuracy during ETL. We seek a

senior professional who can guide our team, which, despite some cynicism, is capable when given clear

instructions.

Our team, consisting of around 40-50 members, including 20 developers and 10 individuals focused on NLP,

recently welcomed a new master;s graduate, bringing fresh expertise to our algorithmic challenges. However, we

lack comprehensive team capabilities. Our preference is for a homogenous team, given the challenges of

managing multinational teams across different time zones.

Our immediate need is not for Amazon or SageMaker experts but for someone to address the core aspects of large

language model application. We consider a hybrid model, identifying an MVP& person for immediate needs,

supported by a flexible structure that accommodates project-specific specialists. This approach is necessitated by

our complex infrastructure, which requires a significant onboarding period to fully grasp.

In summary, we seek a candidate capable of integrating into our system, understanding our sophisticated

processes, and contributing effectively. This involves not just technical skills but also adapting to our

organizational culture and procedural requirements. The ideal candidate would be evaluated over several months,

allowing for a comprehensive assessment of their fit and contribution to our objectives.

Gamito is a licensed recruitment agency under number 1820 / 16.12.2014 with free-of-charge services to the candidates.

High social package, home office, flexible work time.

Guidance and onboarding will be provided.

Regular Bulgarian working hours with flexible time and excellent social package