Krish

Post 1/5 - Intro Just wrapped prompt engineering sprint. 18-20 hrs/week. Shipped 4 deployed tools that solve real workflow pain: Email intent parser, Email tone rewriter, doc generation and Meeting Notes Summarizer. Here's what I built 🧵 GitHub Repo +Live Demo link👇

Email Intent Parser — AI-Powered Support Email Triage Paste a raw customer email → get structured JSON back in 2 seconds: intent, urgency score, tone, and recommended next action. Live demo, code link given in comments...

Day 9 of my prompt engineering sprint: taught an LLM to diagnose customer complaints instead of just answering them. Plain prompt → jumps straight to an answer, sometimes wrong. Add "think step by step" → it reasons through issue → root cause → resolution before committing.

Day 8 prompt engineering: zero-shot vs few-shot Before: classified ambiguous customer messages inconsistently, missed the actual urgency level. After: 4 strategic examples showing the pattern (with edge cases) → locked in correct classification.

Built an extractive QA system using deepset/roberta-base-squad2 It reads your paragraph → finds the exact answer span → gives a confidence score. Added it as Tab 2 to yesterday's summarizer app. One app. Two NLP tools. Live on Streamlit. 🔗Demo Link below

I spent a day building a neural network. A 30-year-old algorithm beat it. 5,500 rows. Classical ML: 97% accuracy, 90% spam recall. Neural net: 94% accuracy, 60% recall. More data = neural nets win. Less data = they embarrass themselves. Knowing the difference is the skill.