Generative AI Solution Engineer

Job Details 

Hours: Full Time 

Location: Fully remote. This opportunity is open to you if you can legally work remotely from one of our home countries in the USA, the UK, Sweden, Malta, Hungary, Spain, Poland, or the Philippines. 

Contract: Permanent 

At Play’n GO we enjoy a flexible freedom to shape our own days. Everyone we hire is unique and every role we fill has certain performance expectations. The day-to-day challenges of work and life will be unique to you but our commitment to helping you find a balance will never vary.  

From the GO 

At Play’n GO our purpose is to entertain. Our pleasure is in working with people ready to make a difference that lasts. We are pioneers. We were the first entertainment supplier to recognise the potential of gaming on the GO and create the innovative mobile games which supercharged our industry and inspired our name. Today, Play’n GO is the leading gaming entertainment supplier to casinos in regulated markets the world over. And we are still growing.  

With growth comes opportunity. This could be yours. Are you ready to help us face challenges and find solutions that get us both to the next level of excellence?  

What the role entails:

Some of the key responsibilities of our Generative AI Solution Engineer will be: 

• Data collection and preprocessing
• Identify all data sources required by the system, including documents, databases, web pages etc
• Define data requirements for data engineers who will set up the proper data pipelines (including using Bedrock’s action groups or function calling to fetch real-time data).
• Ensure the data is clean and free from errors and irrelevant information. Enrich data with enough background information for the LLMs to learn.
• Ensure proper semantic representations of (numeric) tables, so an LLM can "see" the structure – e.g., use models such as Amazon Textract to turn tabular data into semantic HTML or Markdown format that LLMs can understand. Further consider extra representation engineering if dealing with long tables to help extract the correct semantic structure for LLMs.
• Extract meta data when needed – e.g., based on LLMs or regex-based custom code.

• Data chunking:
  • Consider the nature of the data (e.g., structured, semi-structured or unstructured; whether tabular; whether numeric), use case requirements (whether precise answer is required) and system performance (quality vs. speed vs. running cost).
  • Then choose the best chunking approach (e.g., semantic chunking, fixed-length chunking, sliding window chunking, hierarchical chunking, chunking based on custom code etc.) accordingly.

• Creating and storing knowledge embeddings:
  • Explore different embedding models (e.g., AWS Titan variants) on Bedrock and select the best performing one.
  • Work with DB engineer to choose the right DB (e.g., Amazon OpenSearch, Kendra, Pinecone or Aurora) to store the knowledge embeddings, judging by the trade-off between quality, speed and running cost.

• User queries processing:
  • Figure out how to best transform the queries prior to retrieval – such as customizing Bedrock’s default prompt template for guided query decomposition, and making use of simple LLMs to ask user for missing information if needed via customizing the prompt template for example.

• Query embedding and information retrieving:
  • Convert transformed queries into an embedding using the same model as in the knowledge embedding step.
  • Then retrieve the most relevant vectorized knowledge chunks, based on similarity search.
  • Filter on metadata if needed.

• Post-retrieval processing:
  • Figure out how to best use simple summarizing LLMs and third-party ranking models (e.g., bge-reranker-large from Hugging Face) to summarize or re-rank the retrieved information.

• Generating Responses: 
  • Feed the combined queries and information chunks to LLMs, to generate the right response – the GenAI engineer needs to choose the best generative LLM (AWS Titan models, Antropic Claude models, Jurassic and Jamba models, Meta Llama models, or Mistral models etc.) from Bedrock based on the manual documentation and experimentation.
  • Implement the mathematical aggregation (via arithmetic) of multiple cells required by data analysis via AWS Bedrock’s action groups or function calling (set up as a dedicated calculator tool for reliable results) which are called during content generation by LLMs.
  • Ensure mandatory semantics rules (such as “the paylines must be within xxx range”) is enforced on LLM via prompt engineering in the prompt template for example, which triggers certain retrieval and/ or generation behaviour (e.g. metadata filtering and/ or adjust the generation content to fulfil the rules).

• Post-generation processing:
  • Figure out how to best implement response quality checking (e.g., use Bedrock’s action groups or function calling to programmatically examine whether certain required information is generated, and further trigger LLMs to request users for input if needed via customizing prompt template for example). 
  • Maintain the history of a conversation if needed (e.g. via Converse API), so users can submit adjustment requirements via a dialogue manner.

What we ask of you:

• Experience of designing, developing and implementing LLMs based conversational AI products in cloud platforms, such as AWS Bedrock.
• Hands-on experience of RAG, LLM fine-tuning, advanced prompt engineering and handling LLM hallucinations
• Strong technical proficiency with Python, SQL and AWS. Able to write production quality code and have adequate version control skills.
• Strong understanding of DevOps practices and CI/CD pipelines.
• Experience of closely working with data scientists and data engineers.

If you feel great, you’ll do great. Our employees know that their package of benefits, activities, and initiatives are some of the most generous in the industry. From challenges to chill outs, the fun we share at play helps inspire excellence from the go.  

Apply now to find out more. It’s your best move. This opportunity will remain open until we meet the right candidate.  

GO for it. 

We believe the way forward relies on well-regulated gaming and player safety. Our commitment to this sets the standard for our industry.