Software Engineer, LLM Pipeline & AI Agents

**Overview** TSOLife's production LLM pipeline processes audio interviews between senior living communities and their residents — transcribing and diarizing the audio, indexing it for retrieval, and using RAG to answer 85\+ structured questions about each resident's history, preferences, and well\-being. This data builds a rich, personalized profile for every resident, helping communities provide better, more tailored care and programming. We're looking for a backend engineer who can operate and scale this production RAG system, while also designing and building stateful AI agents using LangGraph for a related project. You will own the full lifecycle of these AI features: from prompt design and model routing to operating the FastAPI and Celery backend services on Amazon EKS. What You'll Do Day\-to\-Day * **Operate \& Scale Services:** Ship, maintain, and monitor a production FastAPI \+ Celery backend service running on Amazon EKS. * **Optimize the RAG Pipeline:** Continuously improve dialogue chunking, vector retrieval (Aurora pgvector), context assembly, and prompt design to ensure strict JSON adherence and high faithfulness. * **Build AI Agents:** Design and maintain stateful AI agents using LangGraph for a sibling product initiative. * **Evaluate \& Benchmark:** Compare prompts, retrieval strategies, and models. You'll run consensus\-style accuracy benchmarks and safely roll out per\-question routing changes. * **Manage Audio Processing:** Maintain background jobs (Celery \+ Redis) for transcription submission and handle webhook post\-processing from our speech\-to\-text/diarization provider (Pyannote AI). * **Work with Cloud \& Data:** Query MariaDB for reviewer feedback and training data, and manage pipeline document storage using Amazon Aurora PostgreSQL. Required Experience **Core Backend \& ML Stack:** * **Python Ecosystem:** 3\+ years of production experience (Python 3\.12, FastAPI, Uvicorn, Pydantic). * **LLM \& RAG Tooling:** Hands\-on experience with langchain, langchain\-aws, and **LangGraph**. Deep understanding of chunking, vector retrieval, and grounding/faithfulness checks. * **Model Providers:** Experience building with Amazon Bedrock, specifically Anthropic's Claude (Sonnet/Haiku) and Cohere Embed for dual\-task embeddings. * **Background Processing:** Strong experience with Celery and Redis for asynchronous task processing, webhooks, and job correlation. * **Databases:** Proficiency in SQL. Experience with PostgreSQL (specifically pgvector and psycopg), MariaDB, and schema design. **Cloud \& Infrastructure (AWS):** * **Containers:** Docker and Kubernetes (Amazon EKS), including API \+ worker deployments, HPA, load balancers, and TLS. * **AWS Services:** Amazon Aurora PostgreSQL, S3, ElastiCache (Redis), CloudWatch, and ECR. * **Security \& Ops:** Comfort with AWS IAM, Secrets Manager, IRSA (IAM Roles for Service Accounts), VPC networking, security groups, and day\-to\-day kubectl / aws CLI operations. * **CI/CD:** Git