The latest articles on DEV Community by laya reddy (@laya2211). https://dev.to/laya2211 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F1739815%2Fa0d274cd-2817-458c-854a-6910bea37600.png https://dev.to/laya2211 en laya reddy Mon, 12 Oct 2026 03:33:37 +0000 https://dev.to/laya2211/sprouttime-the-live-data-driven-garden-agent-that-gets-you-outside-3don https://dev.to/laya2211/sprouttime-the-live-data-driven-garden-agent-that-gets-you-outside-3don <p><em>This is a submission for the <a href="https://dev.to/challenges/hacktoberfest-week1-2026-10-05">Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass</a></em></p> <h2> What I Built </h2> <p>I built <strong>SproutTime</strong>, an intelligent, data-driven gardening agent designed to maximize your time in nature and minimize digital screen interactions. </p> <p>Instead of trapping users in infinite scroll feeds or complex setups, SproutTime dynamically tracks real-time meteorological conditions for any global city and pipes that context into a local open-weight model. It instantly generates a physical, printable three-tier markdown checklist tailored to the user's exact gardening experience level. It is built for absolute beginners, hobbyists, and experts who want actionable plant tasks so they can turn off their screens immediately and step into the dirt.</p> <h2> Demo </h2> <p>Interact with my live deployed application here:<br> 👉 <a href="https://huggingface.co/spaces/laya-03/garden-planner" rel="noopener noreferrer">https://huggingface.co/spaces/laya-03/garden-planner</a></p> <h2> Code </h2> <p>You can view the source files, UI layout configurations, and dependency definitions directly on the Hugging Face Space repository:<br> 👉 <a href="https://huggingface.co/spaces/laya-03/garden-planner/tree/main" rel="noopener noreferrer">https://huggingface.co/spaces/laya-03/garden-planner/tree/main</a></p> <h2> How I Built It </h2> <p>SproutTime relies entirely on an open-source, local-first architecture:</p> <ul> <li> <strong>Open Data Orchestration (RAG):</strong> The application uses <code>geopy</code> (connected to OpenStreetMap) to securely convert city strings into global coordinates. It then calls the open-source <code>Open-Meteo API</code> to grab real-time climate telemetry (Temperature, Humidity, Rain) without encountering closed data firewalls.</li> <li> <strong>Local Inference Pipeline:</strong> Powered by the open-weight <strong>Qwen-2.5-0.5B-Instruct</strong> model running completely contained inside the container space via a Hugging Face <code>transformers</code> hardware pipeline (<code>torch</code> + <code>accelerate</code>).</li> <li> <strong>Deterministic Few-Shot Anchor Guardrails:</strong> Because lightweight 0.5B models are highly sensitive, I built a rigid conversational structural template block using specialized tokenizer system markers (<code>&lt;|im_start|&gt;</code> and <code>&lt;|im_end|&gt;</code>). This forces the pipeline to deliver pure agricultural value and eliminates structural text clutter.</li> <li> <strong>Lightweight UI:</strong> Built natively with the <strong>Gradio</strong> framework for smooth user interaction.</li> </ul> <h2> Why Does Open Innovation Matter? </h2> <h2> My Agent Session </h2> <p>Open innovation was the absolute foundation of this project. A closed-source API ecosystem would introduce high request latency, cost walls, and unpredictable pricing matrices for casual outdoor hobbyists. </p> <p>By leveraging open-weight structures like <code>Qwen-2.5</code>, I was able to download, optimize, and deploy an entire text generation pipeline directly inside a self-contained container memory environment. This architecture allows developers to build systems that work without relying on third-party cloud data brokers, perfectly aligning with a private, offline "Touch Grass" philosophy.</p> <h2> Prize Categories </h2> <ul> <li> <strong>Main Challenge Category:</strong> Week 1 Challenge - Touch Grass</li> </ul> devchallenge hf26challenge laya reddy Mon, 05 Oct 2026 04:39:18 +0000 https://dev.to/laya2211/build-for-a-friend-an-interactive-dsa-and-algorithm-coach-for-competitive-exam-prep-5e97 https://dev.to/laya2211/build-for-a-friend-an-interactive-dsa-and-algorithm-coach-for-competitive-exam-prep-5e97 <h2> <strong>What I Built</strong> </h2> <p>I built the <strong>GATE CS Visual Dry-Run Tracer</strong> for a close friend who is aggressively preparing for highly competitive core engineering exams like GATE CS, ISRO, and DRDO. </p> <p>They kept losing critical marks on technical algorithm evaluation papers because standard compilers only print out a final code output—they don't show you <em>where</em> your manual dry-run logic diverged at step 3 or 4 of a 10-step execution trace. </p> <p>This app breaks down any complex code block, nested loop, or graph algorithm problem into an interactive step-by-step matrix table. It allows the student to commit their intermediate checkpoint values in a scratchpad before revealing the absolute step-by-step truth, making it a perfect tool for diagnosing logical slips under exam time pressure.</p> <h2> <strong>Demo</strong> </h2> <p>👉 <a href="https://huggingface.co/spaces/laya-03/gate-cs-tracker" rel="noopener noreferrer">Launch the GATE CS Visual Dry-Run Tracer App</a></p> <h2> <strong>Code</strong> </h2> <p>🔗 <a href="https://huggingface.co/spaces/laya-03/gate-cs-tracker/tree/main" rel="noopener noreferrer">View the Application Source Repository</a></p> <h2> <strong>How I Built It</strong> </h2> <p>The application is built completely on top of an open-source architecture:</p> <ul> <li> <strong>The Core Brain:</strong> We utilize the open-weights <strong>GPT-OSS-20b</strong> model, processed through a high-speed open-access Groq API client pipeline.</li> <li> <strong>Interface &amp; UX:</strong> Designed a clean, distraction-free grid-split interface using the open-source <strong>Gradio</strong> framework.</li> <li> <strong>Deployment Infrastructure:</strong> Deployed as a standalone application inside Hugging Face Spaces utilizing the open-access <strong>ZeroGPU</strong> compute tier wrapper.</li> <li> <strong>Prompt Harness:</strong> Engineered custom low-temperature (<code>temperature=0.2</code>) instruction sets to enforce strict, predictable mathematical markdown transformations from natural text input into deterministic tables.</li> </ul> <h2> Why Does Open Innovation Matter? </h2> <p>Open innovation was the only reason this project was possible to ship over a weekend without a corporate budget:</p> <ol> <li> <strong>Infinite Execution Token Capacity:</strong> Tracing complex algorithm loops step-by-step or detailing matrix transitions consumes thousands of context tokens per query. Standard commercial pay-per-token APIs would bankrupt a student using it for daily prep. Open weights on open-access endpoints mean <strong>infinite free dry-runs</strong>.</li> <li> <strong>Deterministic Stability:</strong> Proprietary models frequently drift in behavior overnight due to black-box alignment updates. By grounding our tool in an open-weights standard with zero temperature variations, we enforce static academic formatting required for error-free textbook analysis.</li> </ol> <h2> Prize Categories </h2> <ul> <li><strong>Open Source AI Novelty</strong></li> <li><strong>Build for a Friend</strong></li> </ul> devchallenge weekendchallenge hf26challenge