AI/ML LLM Engineer

Payment: Weekly worksheet submitted Friday · reviewed Monday · invoice paid within 7 days of acceptance What is Ladybird? A Manchester healthtech startup. Small team. Moving fast. Building inCall - an AI receptionist for NHS GP practices. Project-based engagement with option to extend. What is this role? Hands-on AI and LLM engineering. You will integrate OpenAI as an interim model from day one, then design, build, fine-tune, and deploy a LLaMA 3 8B model from scratch - replacing OpenAI in the live pipeline by end of Month 3. Two tracks running simultaneously. Requirement, read this first You must have production experience fine-tuning and deploying large language models - not just calling APIs. Track A Day one OpenAI GPT integrated into the live 3CX call pipeline via AWS Transcribe and ElevenLabs so the full voice flow is testable immediately. Track B Months 1 to 3 LLaMA 3 8B fine-tuned on 10,000+ synthetic NHS GP call examples using LoRA on SageMaker. Deployed to EC2 g4dn.xlarge. Replaces OpenAI by end of Month 3. If you have never fine-tuned a model and deployed it to a production inference endpoint, this role is not for you. What you will build OpenAI GPT integrated into live call pipeline from day one AWS Transcribe (en-GB) with custom NHS GP vocabulary ElevenLabs TTS for voice responses 10,000+ synthetic NHS GP call training dataset LLaMA 3 8B fine-tuned via LoRA on SageMaker SageMaker inference endpoint - minimum 70% accuracy on held-out test set NHS PDS FHIR API integrated for patient context (Month 3) LLaMA replaces OpenAI in live pipeline (Month 3) Safety and fallback layer implemented and documented 3-month delivery plan Month 1: OpenAI integrated into live pipeline. Full voice flow demonstrated. Synthetic dataset design approved. 1,000+ example outlines complete. Month 2: LLaMA 3 8B fine-tuned on 10,000+ examples. SageMaker endpoint live. Minimum 70% accuracy on 100-example held-out test set accepted in writing. Month 3: NHS PDS FHIR integrated. LLaMA replaces OpenAI in live pipeline. Full POC demo. All eight NHS GP call scenario types demonstrated. Safety layer confirmed.

• Production LLM fine-tuning, LoRA, QLoRA, or equivalent
• AWS SageMaker, training jobs, endpoint deployment, evaluation
• Synthetic data generation for model training
• Understanding of NHS or healthcare AI safety constraints
• Ability to enforce no real patient data in training architecturally
• AWS Transcribe and ElevenLabs or equivalent STT/TTS
• OpenAI API integration
• Ability to evaluate model output for production readiness

Useful but not essential NHS FHIR API · LangChain · RLHF · DPO · SageMaker Ground Truth · Hugging Face · DCB0129 awareness Not suitable if

• You have only called LLM APIs rather than fine-tuned and deployed models
• You cannot architect a patient data firewall structurally - not just as a policy
• You are not comfortable owning model evaluation and sign-off independently

Complete Form: https://ladybirdltd.jp.larksuite.com/share/base/form/shrjpwYz1RbzWxvm4rVRRlkYngf Video: 60-120 second intro video. 3 questions, full brief provided Group interview: 90 minutes via Google Meet You will hear back within 48 hours at every stage either way. Application question(s):

• This is a fixed-term independent contractor engagement lasting three calendar months, with the possibility of extension by mutual written agreement only. Are you aware of and comfortable with this?
• The fee for this engagement is £500 per calendar month inclusive of VAT, totalling £1,500 for the full initial term. Are you aware of this rate?
• This role requires your full commitment for the duration of the engagement. You will attend a daily standup at 12:00 GMT/BST Monday to Friday, submit a weekly worksheet every Friday, and deliver agreed monthly outputs on time.Can you commit to this as your primary engagement for the full three-month term?