Doclium
PDF and document tools plus AI Q&A with citations: merge, split, compress, convert, OCR, and structured extraction workflows.
AI, Document intelligence, SaaS
AI Mobile App Development for Flutter and iOS
I build AI products and connect useful model features to mobile workflows. The starting point is the task a person needs to finish, the data they can share, and the behavior when the model gets something wrong.
Discuss your AI productA useful feature has a clear boundary
Doclium combines document tools with AI questions and citations; TopLLM helps people compare models. These are my own products, separate from the client Flutter portfolio. For a mobile integration, define what the model may suggest and which actions still require explicit user confirmation.
Design the waiting and failure states
Streaming can make a long response easier to follow, but the app still needs cancellation, retry limits, and recovery after a lost connection. Put provider credentials on the backend. Set a per-request budget and make timeouts visible instead of leaving a spinner running.
Ground answers and keep providers replaceable
For retrieval over private documents, access checks must happen before retrieval. Preserve source references with the answer. Keep model-specific code behind a small interface so a provider change does not spread through the mobile UI. Evaluate representative examples before comparing model costs.
When AI should not be in the app
A deterministic calculation, a fixed business rule, or a simple lookup usually needs ordinary code. On-device inference is a separate decision involving device support, model size, and privacy requirements. Voice and image features should follow a concrete user need rather than expand the scope by default.
Related product work
TopLLM
Helps users compare LLMs and choose the right model for coding, reasoning, learning, images, and cost, with plain-English recommendations over benchmarks.
AI product, LLM evaluation, Web
Before we start
Can AI be added to an existing Flutter product?
Yes, when the app and backend provide a suitable integration boundary. Start with one workflow, a small evaluation set, and a fallback before expanding the feature.
Do you train a custom model for every project?
No. Start by testing whether an existing model API or conventional code solves the problem. Custom training needs a separate data, cost, and maintenance case.