Computer vision, RAG pipelines, voice agents and document AI. Not demo AI but systems that handle production traffic month after month. Claude 4.7, GPT-4.5, Mistral and local models.
Book an intro →
AI development at NedDev runs on a mix of Claude 4.7, GPT-4.5 and local models, embedded in a RAG pipeline on a vector database. The price of a production AI feature is fixed after a short intake, and the feature goes live in 4-8 weeks. We keep the models up to date and monitor prompt drift via Sentry and custom dashboards.
AI development is the right choice when you want to automatically process large volumes of documents, conversations, or images: document analysis, voice agents, computer vision, and RAG systems built on your own knowledge base.
Our own base for multi-tenant SaaS: a Laravel + Filament starter, audit trail and tenant management. We build client platforms on it.
Used internally EDITORIAL · LONG-FORM ↗What we write down as we learn it. Case studies, technical write-ups, design decisions. No content marketing, just real knowledge.
42 articles RESEARCH · AI EXPERIMENTS ↗Side projects and R&D. Voice-agent prototypes, RAG pipelines, AI knowledge-graph experiments. Some become products. The rest teach us something.
455 examples in the LabAt NedDev a production AI feature depends on data quality, integrations and the validation work involved. RAG on your own documents takes more work, including vector-database setup, because the documents have to be indexed and made searchable. Computer vision needs the most preparation, as image models require more training data and careful validation before they run reliably in production. After a short intake you get a fixed price.
We combine Claude 4.7 Sonnet for long-context tasks, GPT-4.5 for reasoning, Mistral for on-premise scenarios and Grok 4.3 for real-time data. For computer vision we use our own YOLO + Shapely pipelines.
Every production AI at NedDev has prompt versioning, output validation via JSON schema, fallback models and monitoring on latency, cost and output quality via Sentry. We log every generation so regressions stay traceable.
RAG stands for Retrieval Augmented Generation. The system first finds relevant chunks from your documents via a vector database, passes those as context to an LLM, and generates an answer with source citations. That is how Lexi AI handles labor-agreement questions.
