← All Timeline
experience

Software Engineer — Microsoft at Microsoft — Purdue Data Mine

Aug 2024 — Dec 2024 · West Lafayette, IN

Multi-source social media sentiment pipeline PoC over 1,000 images & 240 videos from 3 sources. Benchmarked 4 LLMs vs RoBERTa baseline (+40% insight capture), reduced analysis time 1–2 h/cycle, added auto-translate to widen dataset.

Highlights

  • Led 5-member team through 8-sprint agile cycle; owned 30-item backlog & documentation to deployment; delivered PoC on schedule
  • Designed end-to-end sentiment analysis pipeline for Microsoft game update response — processing 1,000 images & 240 videos from 3 media sources, boosting insight accuracy 40%
  • Improved insight capture 40% over RoBERTa baseline by benchmarking/selecting across 4 LLMs & small language models reading emoji and mixed-mode signals; PySpark pipeline
  • Automated contextual analysis deploying 4 distinct LLMs; added auto-translate feature to increase analysable dataset
  • Cut analysis time estimated 1–2 hours per cycle
  • Tools: Python, Databricks, PySpark, SQL, LLMs, MS Project
Agentic AIAI DeliveryData Platform

Related Projects