Software Engineer, Beacon AI

• The ideal candidate is hands-on, comfortable with ambiguity, and excited to build from first principles
• End-to-End Ownership: Drives work from design through production, including on-call and continuous improvement
• LLM Systems Experience: Shipped or operated LLM-powered applications in production. Familiar with RAG design, prompt versioning, and chain orchestration using LangChain or similar
• AWS Depth: Strong with core AWS services such as VPC, IAM, KMS, CloudWatch, S3, ECS/EKS, Lambda, Step Functions, Bedrock, and SageMaker
• Data Engineering Skills: Comfortable building ingestion and transformation pipelines in Python. Familiar with Glue, Athena, and event-driven patterns using EventBridge and SQS
• Security Mindset: Applies least privilege, secrets management, network isolation, and compliance practices appropriate to sensitive data
• Evaluation and Metrics: Uses quantitative evals, A/B testing, and live metrics to guide improvements
• Clear Communication: Explains tradeoffs and aligns partners across product, security, and application engineering

Desirable

• We are seeking skilled Cloud and ML Infrastructure Engineers to lead the buildout of our AWS foundation and our LLM platform
• You will design, implement, and operate services that are scalable, reliable, and secure
• The broad scope means focus areas in LLM/ML Infra and IoT infra are strong bonus points
• For ML Infra, build the stack that powers retrieval-augmented generation and application workflows built with frameworks like LangChain
• Experience with IoT AWS services is a plus
• You will work closely with other engineers and product management
• Design, provision, and maintain AWS infrastructure using IaC tools such as AWS CDK or Terraform
• Build CI/CD and testing for apps, infra, and ML pipelines using GitHub Actions, CodeBuild, and CodePipeline
• Operate secure networking with VPCs, PrivateLink, and VPC endpoints. Manage IAM, KMS, Secrets Manager, and audit logging
• Stand up and operate model endpoints using AWS Bedrock and/or SageMaker; evaluate when to use ECS/EKS, Lambda, or Batch for inference jobs
• Build and maintain application services that call LLMs through clean APIs, with streaming, batching, and backoff strategies
• Implement prompt and tool execution flows with LangChain or similar, including agent tools and function calling
• Design chunking and embedding pipelines for documents, time series, and multimedia. Orchestrate with Step Functions or Airflow
• Operate vector search using OpenSearch Serverless, Aurora PostgreSQL with pgvector, or Pinecone. Tune recall, latency, and cost
• Build and maintain knowledge bases and data syncs from S3, Aurora, DynamoDB, and external sources
• Create offline and online eval harnesses for prompts, retrievers, and chains. Track quality, latency, and regression risk
• Instrument model and app telemetry with CloudWatch and OpenTelemetry. Build token usage and cost dashboards with budgets and alerts
• Add guardrails, rate limits, fallbacks, and provider routing for resilience
• Implement PII detection and redaction, access controls, content filters, and human-in-the-loop review where needed
• Use Bedrock Guardrails or policy services to enforce safety standards. Maintain audit trails for regulated environments
• Build ingestion and processing pipelines for structured, unstructured, and multimedia data. Ensure integrity, lineage, and cataloging with Glue and Lake Formation
• Optimize bulk data movement and storage in S3, Glacier, and tiered storage. Use Athena for ad-hoc analysis
• Manage infrastructure that deploys to and communicates with edge devices. Support secure messaging, identity, and over-the-air updates
• Partner with product and application teams to integrate retrieval services, embeddings, and LLM chains into user-facing features
• Provide expert troubleshooting for cloud and ML services with an emphasis on uptime and performance
• Tune retrieval quality, context window use, and caching with Redis or Bedrock Knowledge Bases
• Optimize inference with model selection, quantization where applicable, GPU/CPU instance choices, and autoscaling strategies