Building on AI - Delhi meet-up
summary
Team scrapes 400,000 candidate reviews, manually annotates 10,000 reviews to define 35 signals and builds a pipeline that extracts audio, video and transcript features to detect webcam usage, noise, turn-taking and question topics; evaluation labels 200 interview recordings and appears in an IUI paper; separate work runs pilots of a scripted digital interviewer (RAG/GPT variants), and shares engineering rules: output validation, DDOS rate limits, streaming, semantic caching, Cohere re-rank, Microsoft Presidio redaction, retries (tenacity/async-retry) and forward-compatible model proxies.
AI-written summary · about
chapters (7)
- 0:00Delhi AI meet-up intro
- 2:11AI transforming HR interviews
- 20:36Future of AI interviewing
- 27:40Seven learnings in LLM apps
- 52:23Panel on AI ROI
- 1:31:38Build own LLMs?
- 2:00:00Closing remarks
related talks
01h 03mIt's All About AI
17mAI/ML demo derby
50mBuilding AI ready applications
24mAI and Web Development: Hype or Reality
07mInterview
14mAI for a Billion Voices: Taking AI where there is Maximum Impact
34mI, Human
38mEthical Artificial Intelligence to Advance Welcoming Communities and Identify Potential Communica...