Ahmad ZushanAI · Systems · Platform

Home/Work/№ 19

19 · LLM & Agents

Early NLP & Data Consulting (2024)

Earlier applied-NLP engagements that show a multi-year track record in LLM system design well before the 2025–26 platforms.

Period2024StatusFoundationsAreaLLM & Agents

Highlights

  1. Staged, cost-aware text-to-SQL: identify relevant tables from the top-layer schema first, then send detailed schema only for those, with a clarification loop.
  2. Call-transcript problem-type extraction with Llama 2 and a TF-IDF matrix for partial product-model keywords.
  3. Retail sales forecasting design (LightGBM quantile, TFT/N-HiTS, promo uplift and elasticity); small OpenAI-backed agent tools.