Developer Builds PDF Chat API in One Day Using FastAPI, Gemini, and Qdrant

·2 views

A developer built a fully functional PDF Chat API in a single day, allowing users to upload PDF documents and ask questions about their content using natural language. The system uses a Retrieval Augmented Generation (RAG) architecture, combining Google Gemini for embeddings and chat, Qdrant as a vector database, and LangChain for pipeline orchestration. During ingestion, PDF text is split into chunks, converted into 3072-dimensional vectors via Gemini, and stored in Qdrant; at query time, the most relevant chunks are retrieved and passed to Gemini to generate an answer. The project includes a REST API with authentication, a plain HTML/CSS web interface, and is deployable on Qdrant's free cloud tier. The developer plans to add multi-user support, Docker deployment, and compatibility with other document formats such as Word and Excel.

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)

Missing readOnly Flag in Spring Transactions Cut Connection Pool Efficiency in Half

A development team discovered that several read-only API endpoints — including category listings and dashboard summaries — were annotated with plain @Transactional, omitting the readOnly = true parameter. Without this flag, Hibernate performed unnecessary dirty checks and flush operations on every request, holding database connections open longer than required. At high traffic volumes, this overhead caused measurable connection pool congestion, forcing incoming requests to wait longer for available connections. Adding readOnly = true to the affected methods took a single afternoon and caused peak connection pool wait times to drop by approximately 50%. The incident highlighted how a seemingly minor annotation attribute can have significant performance consequences at scale.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Topic-Specific LoRA Adapters Show Clear Specialization Gains in Small NLP Test

A small experiment tested whether two topic-specific LoRA adapters — one trained on guitar content, one on sourdough — could outperform a base language model on their respective domains. Results showed each adapter achieved its lowest perplexity on its own topic, confirming the intended specialization effect. Cross-topic performance also improved over the base model, though by a smaller margin, indicating partial generalization. The run included independent safety and adversarial checks, both of which passed, adding credibility beyond raw perplexity scores. A GPU memory constraint caused minor operational issues, but the overall findings were considered valid and modest in scope.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developers Share AI-Assisted Personal Projects in Growing Community Thread

A developer on DEV Community shared a personal AI-powered app that aggregates fitness data from Garmin and Strava to help structure training plans. The project was built with AI assistance and serves as a practical tool tailored to the creator's athletic routine. The post sparked a broader conversation by inviting other developers to share their own AI-assisted builds. The thread highlights a growing trend of individuals using AI tools to create small, purpose-built applications for personal use.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds AI Multiplayer Quiz Platform Using React, Supabase, and n8n Workflows

A developer built MindArena, an AI-powered quiz platform supporting both solo practice and live multiplayer contests, without using a traditional backend server. The platform lets users generate AI-created quizzes by selecting a topic, difficulty, and question count, then compete in real-time rooms using shareable codes and leaderboards. Instead of a conventional Node.js or Express backend, the developer routed all logic through n8n automation workflows connected to AI services and a Supabase database. A centralized API layer in React consolidates all frontend-to-workflow communication, simplifying maintenance and error handling. The project highlights how modern automation and serverless tools can replace traditional backend infrastructure for full-featured web applications.

0 comments Read more at DEV Community