Posted inAI AWS Machine Learning
Bedrock Model Distillation: Shrink Coding Assistants for Edge Latency
A senior-engineer guide to Bedrock model distillation, with production-ready patterns, trade-offs, and practical examples.
Posted inAI AWS Machine Learning
SageMaker JumpStart vs Bedrock: When Fine-Tuning Beats Prompting and RAG for Coding Assistants
SageMaker JumpStart vs Amazon Bedrock fine-tuning for coding assistants: decision gate, SFT vs RAG, Bedrock-conversation-2024 data, Nova hyperparameters, on-demand deployment and evaluation.
Posted inAI AWS Machine Learning
Bedrock Knowledge Bases Chunking Strategies That Fit Code RAG
Bedrock Knowledge Bases chunking for code RAG: prefer AST-aware symbol chunks, metadata filters, and hit-rate@k evals over fixed 300-token windows that shred functions mid-signature.
Posted inAI AWS Machine Learning
Codebase Embeddings: Refresh Pipelines When Git SHAs Keep Moving
A senior-engineer guide to codebase embeddings, with production-ready patterns, trade-offs, and practical examples.
Posted inAI AWS Machine Learning
Rag Evaluation Aws: Hit Rate, Faithfulness, and Cost Curves
A senior-engineer guide to RAG evaluation AWS, with production-ready patterns, trade-offs, and practical examples.
Posted inAI AWS Machine Learning
AI Code Review Bot IAM: Least-Privilege Secrets and Pipelines
AI code review bot IAM done right: GitHub App tokens, scoped Bedrock InvokeModel, Secrets Manager, and ephemeral jobs that see a PR diff — not monorepo admin PATs.
Posted inAI AWS Machine Learning
OpenSearch vs pgvector vs S3 Vectors for Codebase RAG on AWS
OpenSearch vs pgvector vs S3 Vectors for code RAG on AWS: filtered ANN, pgvector iterative scans, OpenSearch hybrid search with RRF, S3 Vectors metadata pre-filtering, freshness and recall evaluation.
Posted inAI AWS Machine Learning
Amazon Bedrock Prompt Caching and Batch Inference: Cut Coding-Agent Latency and Cost Without Stale Rules
Bedrock prompt caching vs batch inference for coding agents: cachePoint placement, TTL, usage fields, JSONL batch jobs and stale-rule pitfalls.
Posted inAI AWS Machine Learning
AWS Lambda Coding Agent Sandbox: Safe Tool Isolation for LLM Agents
Build an AWS Lambda coding agent sandbox: ephemeral tool Lambdas, tiny IAM, deny-by-default networking, and structured results — so agents can run tests without owning the account.
Posted inAI AWS Machine Learning
Amazon Bedrock Agents (Classic) in Production: Safe Tools, Memory, Guardrails and the Move to AgentCore
Amazon Bedrock Agents Classic in production: action group Lambda validation, requireConfirmation, sessionAttributes, SESSION_SUMMARY memory, guardrail configuration, and migration mapping to Bedrock AgentCore.