Streaming LLM Architecture Patterns: Sources, Done Events, And Observability
Work period: 2025-2026 streaming LLM systems.
Alex Beattie | Senior AI/LLM Engineer
I build production AI products from 0 to 1: Stella at Ourself Health and KiNDD / NDD Resource Navigator. My work spans scalable Django/Python backends, AWS Bedrock, Strands agents, Langfuse observability/evals, RAG pipelines, cost-aware model routing, GraphQL APIs, Flutter-connected mobile platforms, and the operational details that make AI products dependable after launch.
Start here:
- Production AI routing and evals for a healthcare assistant
- Bedrock, pgvector, and RAG lessons from production
- Streaming LLM architecture patterns for sources, done events, and observability
- KiNDD / NDD Resource Navigator project case study
- Unified Connectors MCP Server project
- Geospatial care-navigation for developmental disability services
- AI assistance for a public-interest healthcare navigator
- AI engineering background and focus areas
Recent Posts
Designing Production AI Routing And Evals For A Healthcare Assistant
Work period: 2025-2026 production AI systems.
Building A Geospatial Care-Navigation Platform For Developmental Disability Services
Project period: 2025-2026 KiNDD platform build.
From Commissioned Website To Nonprofit Startup: The KiNDD Technical Journey
Project period: 2025-2026 founder/engineer work.
Lessons From Bedrock, pgvector, And RAG In Production
Work period: 2025-2026 production RAG systems.
Designing AI Assistance For A Public-Interest Healthcare Navigator
Project period: 2025-2026 KiNDD AI/RAG build.
Building a Markdown-First Citation Pipeline for an AI Health Agent
How I replaced inline URL citations with a token-based reference system that separates machine semantics from user-visible text — and why the old approach wa...
HealthKit + Celery on ECS, Part 3: Deploying Workers to Fargate
One Docker image, two ECS services, an SQS queue with IAM auth, and the Django settings crash that blocks every first Celery deploy.
HealthKit + Celery on ECS, Part 2: The Sync Pipeline
A chunk-based import pipeline with SQS, dispatcher tasks that chain through the queue, fingerprint deduplication, and why we track batch state in PostgreSQL ...
HealthKit + Celery on ECS, Part 1: Why Background Processing Changes Everything
You can’t process Apple HealthKit data inline. Here’s why we needed Celery with SQS, how Redis fits in separately for caching, and the architecture that emer...