This is a submission for the Kaggle Benchmarking Challenge
What I Benchmarked
This is an AI-Assistant and a smart one it can work with and without internet connection. It can solve problems, write simple code, summaries your task, give world wide information even in an offline mode. But it has one flaw because it can't give real time (live) information without internet.
I build this for my sister because she needs AI for studies but we can't buy the expensive subscription of AI models and there are too many students who needs AI for different work but they can't afford it so this is for them an AI which works with and without internet connection.
Models Tested
I used the model "gemma3:1b" because my computers ram is too low to run the large models like "gemma4:e2b".
Findings
Zero-Cost Accessibility: The 1-billion parameter Gemma 3 model runs seamlessly within an ~820MB memory footprint on consumer-grade student laptops, proving that high-quality academic help doesn't require expensive server-side GPUs or active API token costs.
- Resilient Routing Effectiveness: Evaluating the hybrid connectivity layer proved that local fallback architectures can successfully maintain core student workflows (summarization, coding) during internet dropouts without throwing application crash errors.
- The Real-Time Trade-off: While the local Gemma 3 execution handles static logic perfectly, the benchmark highlights a strict dependency on the online routing layer for time-sensitive queries (like real-time local updates), mapping out exactly where lightweight local models need hybrid network support.
Top comments (0)