Faculty often have one guest slot a semester and want it to connect to the course rather than float above it. This is the 90-minute lecture I deliver into machine learning, software engineering and applications courses, with the joins to a syllabus marked so a professor can slot it in.
0–15: what a language model is
Prediction over tokens, context windows, why it is fluent and why it is confidently wrong. Two live calls: one that works, one that hallucinates a citation. Join to the course: probability and sequence models.
15–40: retrieval, live
Chunk a document, embed it, retrieve, answer with context — and watch the hallucination disappear. Then a question the document cannot answer, and the model invents one; that motivates the confidence gate. Join to the course: information retrieval, vector spaces, evaluation.
40–55: where fine-tuning fits
Behaviour versus knowledge; adapters on a quantised base; why evaluation must compare base and tuned models on held-out data. Join to the course: optimisation, overfitting, train/test discipline. Based on FineTune Studio, with a real loss curve on screen.
55–75: how production systems stay honest
- Hybrid retrieval and reranking.
- Confidence gates and refusals.
- Citations and structured logs.
- Fallbacks when a provider fails.
Shown as a walk through RAG.NextUpgrad's architecture with the trade-offs named. Join to the course: software engineering, reliability, testing.
75–90: exercise and questions
A take-home aligned to the course: build a ten-question evaluation set for a small document and report retrieval recall, or reproduce the hallucination-then-context demo on a new document. Fifteen minutes of questions, which in my experience run over.
What faculty should send beforehand
The syllabus, the week the lecture lands in, and what students can already do. With that, the joins above are tuned to the room. Departments can arrange this lecture, on-site or remote, through the invite page.
GenAI engineer and AI product builder with 2+ years shipping production-grade AI systems: RAG pipelines, fine-tuned LLMs, hybrid retrieval and multi-modal apps across vision, speech and OCR, architected end to end from ingestion to deployment. Leads engineering for SCULT INDIA's 14-member team, founded the 500+ member TechVerse Enclave community and has mentored 200+ students. Three hackathon first prizes: Changethon 2025 (IIT Roorkee), Product Genesis at Vividhotsava 2025 (CSJMU Kanpur) and BYTEBATTLE (MeetKats).
Open to GenAI roles, hackathon judging, mentorship sessions and guest talks at colleges. Email pranjulrathour41@gmail.com.
