The lab notebook
Field notes on AI engineering
824 articles · page 27 of 28
Ragas for Enterprise RAG: Scaling Evaluation Across Teams
How enterprise teams use Ragas to standardize RAG evaluation across squads, pipelines, and release cycles without slowing shipping.
Ragas Faithfulness Metric Deep Dive: How Claim Verification Works
A technical breakdown of how Ragas Faithfulness decomposes answers into claims and verifies each one against retrieved context.
Ragas for Chatbot Evaluation Beyond Pure RAG Use Cases
Ragas isn't just a RAG scorer — here's how to use its general-purpose and agentic metrics to evaluate any LLM chatbot.
Ragas for Voice-Based RAG Assistants
Learn how to adapt Ragas metrics and pipelines to evaluate voice-based RAG assistants, from transcript noise to latency-aware retrieval scoring.
Ragas Noise Sensitivity Metric: Testing Retrieval Robustness
Learn how Ragas Noise Sensitivity measures whether your RAG pipeline breaks when retrieval pulls in irrelevant chunks.
Ragas Open Source Roadmap: What's Coming Next
A grounded look at where Ragas is headed after its v0.4 rewrite, from the collections API to the open questions the maintainers still haven't settled.
Ragas for Streaming RAG Responses: Evaluating Partial Outputs
Learn how to adapt Ragas metrics to evaluate streaming RAG outputs without waiting for full generation to finish.
Ragas Test Set Size: How Many Examples Do You Actually Need?
A practical, no-fluff breakdown of how many test examples your Ragas evaluation actually needs, and why "more is better" is the wrong instinct.
When Visual AI Builders Stop Scaling: Signs You Need Real Code
Visual AI builders like LangFlow are great until they aren't — here's how to spot the exact moment your team should drop into code.
Building an AI Feature vs Building an AI Product: The Real Difference
An AI feature bolts a model onto software that already works. An AI product falls apart the moment the model does. Here is why that gap matters.
AI Guardrails Explained: Content Filtering and Output Validation
A practical guide to AI guardrails, covering input content filtering, output validation, and the engineering patterns that keep LLM apps safe.
AI Hallucination Explained: Why Models Make Things Up
Understand why large language models confidently invent facts, how hallucination happens under the hood, and the engineering patterns that reduce it.
Batching and Streaming: Two Ways to Serve LLM Requests
Learn how batching and streaming shape LLM serving, when to use each, and how they change latency, throughput, and cost in production.
Chain-of-Thought Prompting Explained: Why It Works
Learn how chain-of-thought prompting makes LLMs reason step by step, why it works, and how to apply it with real prompt examples you can copy today.
Cold Start Problems in AI Products: Why Day One Is the Hardest
Learn why AI products fail on day one and the practical engineering patterns that solve the cold start problem before it kills your launch.
Context Windows Explained: Why Bigger Isn't Always Better
A practical guide to what context windows really do, why huge ones can hurt, and how to build AI systems that stay sharp.
Embeddings Explained: How Text Becomes Vectors
A hands-on guide to how text becomes numbers, why cosine similarity matters, and how embeddings power search, RAG, and recommendations.
Fine-Tuning vs LoRA vs Full Fine-Tuning: What's the Difference
A clear, practical breakdown of fine-tuning, full fine-tuning, and LoRA so you know which technique to reach for and why.
Function Calling Explained: How LLMs Use Tools
A practical guide to how large language models use function calling to invoke real tools, return structured data, and act on the world.
LangSmith Custom Evaluators: Beyond the Built-In Metrics
Learn how to write custom LangSmith evaluators in Python to score LLM outputs on the rules your product actually cares about.
LangSmith Experiment Comparison: Reading the Diff That Matters
Learn how to read LangSmith experiment comparisons so you can tell a real regression from noise and ship prompt changes with confidence.
LangSmith for Multi-Tenant Applications: Isolating Traces Per Customer
Learn how to isolate LangSmith traces per customer using metadata, tags, and projects so every tenant stays cleanly separated in your LLM app.
LangSmith Trace Retention: Managing Storage Costs Over Time
Learn how LangSmith trace retention works and practical tactics to keep observability storage costs predictable as your LLM app scales.
LangSmith for Voice Agent Debugging
Learn how LangSmith traces, evaluations, and datasets turn a black-box voice agent into something you can actually debug and improve.
Latency vs Throughput in LLM Applications: Why Both Matter
Latency and throughput pull LLM apps in opposite directions. Learn what each means, how they trade off, and how to tune both without guessing.
What Is a Mixture of Experts (MoE) Model?
A plain-English guide to Mixture of Experts (MoE) models: how sparse routing, gating networks, and experts make large language models faster and cheaper.
Prompt Injection Explained: The Security Risk Every AI App Faces
Prompt injection explained for AI engineers: how attackers hijack your model through text, and the layered defenses that actually reduce the risk.
Quantization Explained: Running Large Models on Smaller Hardware
Learn how quantization shrinks large language models so they run on smaller GPUs and even laptops, without wrecking accuracy.
Synthetic Data for AI: Generating Training and Test Data
Learn how synthetic data powers AI training and testing, with practical generation methods, code, and quality checks for real projects.
System Prompts vs User Prompts: Understanding the Roles
Learn the real difference between system prompts and user prompts, why the roles matter, and how to structure messages that make LLMs behave.