Things I've shipped.
A collection of SaaS products, developer tools, and open-source projects built over the last five years.
Resume TLM
A modular PyTorch NLP pipeline for structured resume extraction. DistilRoBERTa-Base + GLU spatial fusion + CRF head. 66M parameters, ~78ms CPU inference — no GPT API call at inference time.
Built from scratch: custom HITL labeling workbench, auto-labeling pipeline (Gemma 4 + Ollama), and a 4-stage modular training engine. 203 resumes labeled, 5 model types.
Hybrid99
Hybrid Classical CV + ML pipeline for rice quality grading. MicroUNet segmenter (94.92% IoU) + TinyMobileClassifier (9 classes, 82.45% val acc) targeting ~1.56MB combined INT8. Built with a custom masking tool for pixel-level COCO annotations.
Classical CV handles 80–90% of grains for free. ML only fires for clusters. Built a custom masking annotation tool, synthetic data pipeline, and two production-grade models.
Comprehensive platform for schools, colleges, and libraries. Integrates management tools for administrators, staff, students, and families. Multi-tenant, edge-deployed.
Hotel & restaurant management with modern design and smart reporting. Manage billing, orders, and reports for Restaurant, Bar & Rooms in one place.
Fluent query builder with relationships, migrations, and lifecycle hooks — built specifically for Cloudflare D1 and the edge runtime.
Generate personalized emails and research 6,000+ companies with retrieval-grounded context. AI writes compelling outreach that gets responses.