NeraBlog
Expert analyses, real-world case studies, and the challenges and solutions we tackle daily.
ragevaluationembeddingsHow to Generate a Retrieval Eval Set From Your Own Corpus
Hand-written search test cases measure the queries you imagined, not the ones users type. How to generate ground truth from the documents you already index, the leak filter that keeps the test honest, and the precision@k numbers you get out.
Read article →
prompt cachingLLM inference costcache breakpointsCase Study: Cutting LLM Cost by 36% with Prompt Caching
Prompt caching cut our LLM inference cost by 36% in production. How prompt caching works, where to put cache breakpoints in a long system prompt and tool schema, and the mistakes that quietly destroy your cache hit rate.
Read article →
searchragagentsAI Problems Are Search Problems
Document selection, tool selection, agent memory, intent classification, eval curation — are just a few of the things nobody files under search, all of which are search. What the reframe buys you, and the four moves that follow from it.
Read article →
ReAct agentstool callingHow to Build ReAct Agents in 2026
The ReAct paper is from 2022 and the loop it describes is now a native model feature. What building a reasoning-and-acting agent actually looks like in 2026: tool calling, state, failure handling, and when the loop is the wrong shape entirely.
Read article →
searchragembeddingsHybrid Search: BM25, Semantic Search or the Best of Both Worlds
BM25 nails the exact terms your users type. Vectors catch what they meant. Where each one fails, how fusion fixes it, and why your agent's search_products tool hides thirty years of work behind one function name.
Read article →
modelscostUnknown Unknowns: Should I Actually Pay for Fable 5
Starting July 12th, Fable 5 leaves your Claude subscription and goes pay-as-you-go - at twice the price of Opus. Before your auto-mode scripts drain your wallet, here's how to actually decide when the mega-model is worth it.
Read article →
essaybusinessHow I Learned to Stop Worrying and Love AI
Strip away the marketing hype and the doomsday tweets, and you find that enterprise tech is still incredibly messy. The models are powerful, but they are not magic. Here is how I traded the dread of the exponential curve for the work in front of me today.
Read article →
ragcustomer supporttutorialBuilding AI Support Agents: Question Extraction as a Retrieval Strategy
Users ask questions; docs contain answers. Embedding both and hoping they match is why support RAG misses. Index the questions each document answers instead.
Read article →
evaluationagentsMeasure What Counts: The AI Engineering Approach to Agent Evaluation
Stakeholders love single numbers. Oh boy they do.
Read article →
essaybusinessThe Changing Corporate Order: How AI is Reshaping the Business Landscape
Throughout history, dominant powers-nations or corporations-failed to recognize the warning signs until it was too late. Ray Dalio’s The Changing World Order illustrates how…
Read article →
voiceagentscustomer supportAI Voice Agents: When Your Best Customer Service Rep Isn't Human
How AI-powered voice automation is transforming customer experience - no hold music required
Read article →
evaluationproductionBuilding Confidence in AI Applications: A Methodology for Iterative Development
As AI applications move from prototypes to production systems, one question becomes increasingly critical: how can we be confident that each iteration actually improves our…
Read article →
agentsdatabasesenterpriseFrom Code to Conversation: The AI Agent That Speaks Database
Fortune 500 companies are transforming enterprise data access with AI that turns natural conversation into precise database queries—and it's already reshaping how businesses…
Read article →
case studyagentsenterpriseBuilding Natural Language Interface for Vulcan Cyber ExposureOS
Building the natural-language layer over Vulcan Cyber's ExposureOS: turning a security admin's plain-English question into a query the platform can answer.
Read article →
reactpythontutorialBuilding a Python React Agent Class: A Step-by-Step Guide
In this blog post, we’ll walk through the creation of a React agent class in Python. After going over our previous post (ReAct: Merging Reasoning and Action to Elevate AI Task…
Read article →
reactresearchReAct: Merging Reasoning and Action to Elevate AI Task Solving
“ReAct: Synergizing Reasoning and Acting in language models" is a research paper that explores the use of large language models (LLMs) to generate both reasoning traces and…
Read article →
financebusinessFrom Algorithms to Assets: AI’s Role in Revolutionizing Trading
Nearly every aspect of life feels the relentless march of artificial intelligence (AI), but the financial sector experiences it primarily through the market. Risk capital funds…
Read article →