The latest articles on DEV Community by pybankers (@pybankers). https://dev.to/pybankers 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%2F4044676%2Fd04178c0-45f7-4fec-9dcb-782e74689ab9.jpg https://dev.to/pybankers en pybankers Mon, 26 Oct 2026 00:34:23 +0000 https://dev.to/pybankers/from-430-mb-to-13-mb-building-an-offline-ai-model-for-indian-gst-invoices-7nl https://dev.to/pybankers/from-430-mb-to-13-mb-building-an-offline-ai-model-for-indian-gst-invoices-7nl <p>We built Tiny GST, a compact AI model designed to extract structured data from Indian GST invoices directly on-device.</p> <p>The model:</p> <ul> <li>Runs fully offline</li> <li>Uses only 13.17 MB after INT8 ONNX quantization</li> <li>Achieves 64.2% entity F1 on unseen vendor layouts</li> <li>Reaches 84.6% F1 for unit-price extraction</li> <li>Reaches 80% F1 for HSN/SAC extraction</li> <li>Achieves 100% CGST amount F1</li> <li>Generates validated TallyPrime-formatted XML</li> <li>Includes invoice deduplication using supplier GSTIN, invoice number, and financial year</li> </ul> <p>The goal is simple: reduce cloud costs, protect sensitive invoice data, and make invoice processing work even in warehouses or rural areas with unreliable internet.</p> <p>The project is currently an early developer release. The XML workflow has been tested against a mock Tally server; live TallyPrime import testing is still pending.</p> <p>🔗 GitHub profile: <a href="https://github.com/sanprat" rel="noopener noreferrer">https://github.com/sanprat</a><br> 🔗 Tiny GST repository: <a href="https://github.com/sanprat/GstEngine" rel="noopener noreferrer">https://github.com/sanprat/GstEngine</a><br> 🔗 Full engineering article: <a href="https://pybankers.com/blog/from-430mb-to-13mb-offline-ai-gst-invoices" rel="noopener noreferrer">https://pybankers.com/blog/from-430mb-to-13mb-offline-ai-gst-invoice</a><br><br> 🔗 PyBankers: <a href="https://pybankers.com" rel="noopener noreferrer">https://pybankers.com</a></p> <h1> MachineLearning #Android #OpenSource #India </h1> machinelearning android opensource showdev pybankers Sun, 30 Aug 2026 06:17:09 +0000 https://dev.to/pybankers/one-shared-memory-for-every-agent-in-a-project-over-mcp-fully-local-ollama-qdrant-mem0-jlc https://dev.to/pybankers/one-shared-memory-for-every-agent-in-a-project-over-mcp-fully-local-ollama-qdrant-mem0-jlc <div class="crayons-card c-embed text-styles text-styles--secondary"> <div class="c-embed__content"> <div class="c-embed__cover"> <a href="https://github.com/sanprat/agent-memory" class="c-link align-middle" rel="noopener noreferrer"> <img alt="" src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fopengraph.githubassets.com%2F366f06265fd0399cc3fb665f31ba046359556b232db3b7eb3ce42c01ba6540f6%2Fsanprat%2Fagent-memory" height="600" class="m-0" width="1200"> </a> </div> <div class="c-embed__body"> <h2 class="fs-xl lh-tight"> <a href="https://github.com/sanprat/agent-memory" rel="noopener noreferrer" class="c-link"> GitHub - sanprat/agent-memory: Universal MCP memory layer for AI coding agents (OpenCode, Claude Code, Cursor, Kilo Code). Self-hosted with Mem0 + Qdrant. Seed rules once, all agents remember. · GitHub </a> </h2> <p class="truncate-at-3"> Universal MCP memory layer for AI coding agents (OpenCode, Claude Code, Cursor, Kilo Code). Self-hosted with Mem0 + Qdrant. Seed rules once, all agents remember. - sanprat/agent-memory </p> <div class="color-secondary fs-s flex items-center"> <img alt="favicon" class="c-embed__favicon m-0 mr-2 radius-0" src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fgithub.githubassets.com%2Ffavicons%2Ffavicon.svg" width="32" height="32"> github.com </div> </div> </div> </div> agents ai mcp opensource