Developer builds RAG system in Go using Gemini File Search with no vector database

·1 views

Software developer Maneshwar built a lightweight retrieval-augmented generation (RAG) tool in Go to help engineering teams surface insights from internal documents like postmortems and design reviews. Instead of using a vector database or embedding pipeline, the system relies on Google Gemini's File Search feature, which handles chunking, embedding, and indexing automatically. The architecture requires just one Go binary, a SQLite database, and two model calls per query against a single hosted document store. Storage and query-time embeddings are free on Gemini's free tier, with costs only incurred at indexing time and for context tokens during generation. The developer also found that proper PDF-to-markdown conversion using pymupdf4llm was critical, as clean headings significantly improved the quality of retrieved chunks.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer builds sentence-buffered moderation pipeline to keep LLM streams safe

A developer building Paramour.chat, an LLM-based roleplay platform, has detailed a technique to moderate streamed AI responses without creating noticeable delays for users. The core challenge is that once a token crosses the network, it cannot be retracted, making pre-delivery moderation essential. The solution treats output as a pipeline: one sentence streams to the user while the next is simultaneously generated and checked by a fast content classifier. If the classifier finishes before the current sentence finishes displaying, the next sentence follows seamlessly with no visible pause. The approach relies on both a sufficiently fast writer model and low-latency classifiers to keep the moderation step hidden within the natural display time.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Shares Lessons Learned Building a WordPress CRM Plugin for Real Users

A developer at Bit Code has shared practical insights gained while building Bit CRM, a sales CRM plugin published on the WordPress.org directory. Unlike tutorials, real-world development meant working inside WordPress's admin environment alongside other plugins, none of which the team controlled. The project required extending an existing ecosystem of plugins — Bit Form, Bit Flows, and Bit Integrations — rather than designing systems from scratch, which shifted the developer's approach to reading and respecting established conventions. Managing multiple modules simultaneously, including a client portal, invoice system, and calendar view, was handled through a strict separation of server state and local UI state. The developer concludes that competitive programming and tutorials build useful habits, but shipping software others depend on exposes an entirely different set of challenges that formal education rarely addresses.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

bitquery-go: Open-Source Go SDK Targets Production-Grade Blockchain Data Integration

A developer has released bitquery-go, an MIT-licensed Go SDK designed to simplify working with Bitquery's GraphQL APIs in production environments. The library treats Bitquery's V1 and V2 APIs as distinct clients rather than interchangeable, helping teams avoid silent fallbacks or schema mismatches. It separates HTTP queries from WebSocket subscriptions and distinguishes GraphQL-level errors from transport failures — common failure points in blockchain data pipelines. The SDK targets Go teams building indexers, analytics backends, trading-data pipelines, and blockchain monitoring tools, and requires Go 1.21 or newer. It is available via the standard Go module system at github.com/tigusigalpa/bitquery-go.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

15-Year QA Veteran Reflects on Humility and Fresh Starts at an AI Startup

A QA professional with 15 years of experience — spanning manual testing, automation, and test management — recently joined an AI agent startup after being laid off and reconnecting with a former colleague who was hiring. Despite deep expertise, the writer describes feeling disoriented on day one and spending the first two weeks deliberately suspending judgment to understand the new environment before critiquing it. The startup's development pipeline and slower, more deliberate pace differ markedly from previous roles, and the writer views that measured rhythm as a strength rather than a flaw. Drawing on a fictional character from a personal writing project, the author illustrates how accumulated experience can become a liability when context changes — a lesson that applies equally to AI systems and seasoned professionals. The central takeaway is that pattern recognition built over years can mislead when the underlying system has quietly shifted.

0 comments Read more at DEV Community