Five production AI apps in one year

Flip through 8 slides ↓All carousels

Slides

01 / 08

Five production AI apps in one year, slide 1 of 8 — 01

Five production AI apps in one year, slide 2 of 8 — 02

Five production AI apps in one year, slide 3 of 8 — 03

Five production AI apps in one year, slide 4 of 8 — 04

Five production AI apps in one year, slide 5 of 8 — 05

Five production AI apps in one year, slide 6 of 8 — 06

Five production AI apps in one year, slide 7 of 8 — 07

Five production AI apps in one year, slide 8 of 8 — 08

The caption

In 2026 I designed, built and shipped five production-grade AI applications end to end at NextUpgrad Web Solutions — each one a different modality, each one deployed with Docker, CI/CD and structured logging: RAG.NextUpgrad (hybrid retrieval, confidence gate, multi-provider fallback — 196 tests), FineTune Studio (QLoRA in a browser with live loss — 107 tests), DocuLens AI (deterministic document intelligence on Groq — 44 tests), FaceVision (browser-only face detection, recognition and liveness — 260+ tests) and the OCR & Speech Workspace on Mistral (page-batched OCR, live speech, document-scoped chat — 40 tests, 87% coverage). This deck is the overview: what each one does, what they share, and what shipping five taught me that one never could. I also build tools.scult.in — 15 free tools, 1,200+ prompts and a 50,000-skill library, free for anyone, no signup.

Keep flipping