The latest articles on DEV Community by Devansh Sharma (@devansh_sharma_ds). https://dev.to/devansh_sharma_ds https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3478309%2Fb6ad8e17-8f65-4b04-b6ce-7945fb006bd8.png https://dev.to/devansh_sharma_ds en Devansh Sharma Fri, 01 May 2026 06:58:29 +0000 https://dev.to/devansh_sharma_ds/im-building-a-real-jarvis-in-python-heres-whats-working-and-whats-not-3a99 https://dev.to/devansh_sharma_ds/im-building-a-real-jarvis-in-python-heres-whats-working-and-whats-not-3a99 <p>Most AI tools today are just chat interfaces.</p> <p>You open them → type → get a response → close them.</p> <p>That’s not an assistant.</p> <p>So I started building VISAR Edge — a system-level AI assistant that’s always present, not something you have to “open.”</p> <p>🧠 The Core Idea</p> <p>Instead of:</p> <p>«User → Prompt → Response»</p> <p>I’m building:</p> <p>«Environment → Context → AI → Action»</p> <p>⚙️ Current Architecture</p> <ul> <li>Python (core system)</li> <li>PySide6 (UI layer experiments)</li> <li>SpeechRecognition (voice input)</li> <li>Gemini API (LLM)</li> <li>System hooks (clipboard monitoring, active window tracking)</li> <li>Chroma Vector DB,many more...</li> </ul> <p>🚀 What It Can Do (So Far)</p> <ul> <li>Floating UI overlay (works across apps)</li> <li>Voice-triggered interactions</li> <li>Clipboard-aware responses</li> <li>Modular feature system (plug-and-play architecture)</li> </ul> <p>⚠️ What’s Harder Than Expected</p> <ol> <li>Real-Time Context Awareness</li> </ol> <p>Capturing meaningful context without overwhelming the system is tricky.</p> <p>Too little → useless assistant<br> Too much → performance nightmare</p> <ol> <li>Latency</li> </ol> <p>Cloud APIs are powerful but slow for an “always-on” experience.</p> <p>This is pushing me toward:</p> <ul> <li>Local models</li> <li>Hybrid inference (local + cloud fallback)</li> </ul> <ol> <li>System Design &gt; AI</li> </ol> <p>Big realization:</p> <p>«This is more of a systems engineering problem than an AI problem.»</p> <p>Managing:</p> <ul> <li>State</li> <li>Memory</li> <li>Context flow</li> <li>Event triggers</li> </ul> <p>…is harder than calling an LLM.</p> <p>🧩 What I’m Exploring Next</p> <ul> <li>Local LLM integration (7B–13B range)</li> <li>Vector memory systems</li> <li>Event-driven architecture</li> <li>Background processing optimization</li> </ul> <p>❓ Where I Need Feedback</p> <p>If you’ve built anything similar, I’d love your thoughts:</p> <ul> <li>Is Python a bad long-term choice for this?</li> <li>Best way to handle continuous context streams?</li> <li>How would you design memory for this kind of system?</li> </ul> <p>🔗 Full Project Context</p> <p>I’ve documented the vision, UI, and direction here:<br> 👉 <a href="https://www.linkedin.com/in/devanshsharma987" rel="noopener noreferrer">https://www.linkedin.com/in/devanshsharma987</a></p> <p>💬 Final Thought</p> <p>Everyone is building smarter chatbots.</p> <p>I’m trying to build something you don’t have to think about using.</p> <p>If this fails, it’ll fail by aiming too high — not too low.</p> <p>Would appreciate honest feedback.</p> ai productivity python showdev