local-ai-research-assistant

LocalAI Research Assistant is a local-first AI research platform that ingests webpages, performs grounded research tasks with Ollama models, compares model performance, and exposes the system through CLI, FastAPI, and Gradio interfaces. It is described as an AI engineering portfolio project built around practical local model workflows: source ingestion, structured prompting, Pydantic validation, model comparison, observability, and a browser-based demo UI. The system uses local Ollama integration through an OpenAI-compatible client, with a provider abstraction for model-independent research logic, and supports llama3.2 and gemma3. Webpage ingestion combines static scraping with BeautifulSoup and a Playwright fallback for JavaScript-rendered pages, producing a normalized WebDocument source model. Research tasks include structured summaries, topics, key facts, grounded Q&A, and research reports, with JSON-first model outputs validated by Pydantic and source-grounding prompts that instruct models not to invent facts. The platform also provides multi-model comparison on the same ingested document, deterministic evaluation checks for parsing, grounding, completeness, and latency, and local observability through request IDs, structured logging, local metrics, and a benchmark runner. Interfaces include CLI, FastAPI, and Gradio, with health/readiness endpoints and pytest and Ruff coverage for normal local/CI workflows.