Topic
16 articles found
In my last post, I built a RAG pipeline from scratch — no LangChain, just FastAPI + FAISS. It scored 17/19 on my test set. But two questions failed: "Who is the CEO?" — couldn't find it "How many employees does Zentara have?" — couldn't find it Both answers were right there on page 1. So what went w
TL;DR Converting scraped web content directly into Markdown reduces token consumption by up to 90% while preserving the semantic structure needed by LLMs. Combining Markdown extraction with PostgreSQL and the pgvector extension creates a highly efficient, production-ready Retrieval-Augmented Generat
Hi everyone, I’ve been working on optimizing RAG pipelines and LLM workflows that are specifically designed to process dense, domain-specific academic literature and scientific text. One consistent roadblock I keep encountering is the massive drop in model recall when feeding LLMs text data derived
Three weeks ago I needed a way to query a large document corpus without sending everything to an LLM every time. The answer was a RAG (Retrieval-Augmented Generation) pipeline — but I wanted to build it inside n8n, not a Python script that I'd have to maintain separately. Here's the architecture I l
A backend engineer's first step into AI Engineering: embeddings, vector search, and the chunking bug that made everything click. I have been a backend engineer for a while now: TypeScript, NestJS, distributed systems, APIs in production. I like that work. But at some point I started paying attention
Everyone reaching for a vector database when building RAG is solving the wrong problem first. For most domain-specific corpora — technical documentation, company knowledge bases, article archives — BM25 retrieval is competitive with semantic search, costs a fraction of the compute, and is dramatical
Introduction As enterprise adoption of LLMs grows, inference costs, hallucinations, and retrieval inefficiencies are becoming major production challenges. Traditional vector-based Retrieval-Augmented Generation (RAG) improves grounding, but it still struggles with multi-hop reasoning and relationshi
I Built a RAG Pipeline From Scratch and It Completely Changed How I Think About AI I've been writing code for 3+ years. I thought I understood how AI worked. I didn't. Not until I sat down one weekend, opened a blank Node.js project, and decided to build something I'd been curious about for months —
Most RAG systems get tested with a handful of happy-path questions. Someone asks "what is machine learning?", gets a reasonable answer, and calls it done. Then it goes to production and users find the edge cases, hallucinations on out-of-scope questions, failed refusals on adversarial prompts, laten
The Problem If you are building AI applications with LLMs, you know the pain: raw HTML is useless for training data. You need clean, structured Markdown. Most solutions like Firecrawl or Crawl4AI require setup, dependencies, and often paid plans. You could write your own parser: import re import url
In 2024, 72% of production RAG systems fail to meet p99 latency SLAs of 500ms, per a Gartner study of 1200 enterprise deployments. The root cause? 89% of teams misconfigure vector database integration with orchestration frameworks like LlamaIndex. This deep dive fixes that, with benchmark-backed cod
Your RAG pipeline retrieves real documents — and still hallucinates. Here's the retrieve → generate → verify pattern that catches it before your agent acts, with working Python code you can run right now. Your RAG pipeline retrieves three real documents. The LLM reads them. It generates a response t
I'm building a production-grade, multi-tenant AI support agent in TypeScript. No Python. No LangChain. Just Node.js, PostgreSQL with pgvector, and a lot of wrong assumptions that taught me more than any tutorial. This post covers three things I got wrong while building my RAG (Retrieval-Augmented Ge
I built a production-grade RAG system called PrecisionRAG. It combines Self-RAG and CRAG (Corrective RAG) techniques, runs on LangGraph, has hallucination checking, answer revision loops, usefulness checks, corrective re-retrieval before web search fallback - and more. Then I asked it a simple factu
A 2025 peer-reviewed study (Vectara, NAACL 2025) found something most RAG teams get backwards: Chunking strategy has equal or greater impact on retrieval quality than embedding model selection. Teams spend weeks choosing between OpenAI, Cohere, and Jina embeddings — then split documents every 512 to
A practical guide to caching layers across the RAG pipeline, from query embeddings to full query-response reuse The post Beyond Prompt Caching: 5 More Things You Should Cache in RAG Pipelines appeared first on Towards Data Science.
