RAG Interview Skit Checked: Real Fixes, Free Courses, $89 Community
By Matt Cullerton, founder of Mavric and Apso
A TikTok sketch from @bashifuirkashi pits a "Theory" candidate against a "Builder" in an AI engineer interview about retrieval-augmented generation (RAG). The Builder's fixes for cost, latency and hallucination are all real techniques with documented APIs behind them. The video ends by promoting the BASWE.AI Engineer community on Skool, which costs $89 a month, and the lessons shown in its own screen recordings link to three free public courses.
Is the "Theory" RAG answer wrong?
The Theory candidate says: "first you ingest your documents chunk them create embeddings store them in a vector database retrieve the relevant chunks then pass them into an llm to generate the answer."
That matches the architecture in the original paper, Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks, submitted May 22, 2020 by Patrick Lewis and co-authors. The paper pairs a "parametric memory" (a pre-trained seq2seq model) with a "non-parametric memory" that is "a dense vector index of Wikipedia, accessed with a pre-trained neural retriever."
The answer is incomplete on operations, evaluation and failure isolation, which the Builder adds. The Builder opens with "same foundation," which concedes the basic description is right. The skit ends as a hiring outcome ("you're hired") rather than a ruling on correctness.
How do you scale a RAG pipeline for 10x traffic?
Asked what happens when "traffic just went up by 10x," the Builder lists several moves. Each one maps to documented vendor behavior.
Batch embeddings during ingestion. The OpenAI Batch API gives a 50% discount versus synchronous requests, uses a 24-hour completion window, supports /v1/embeddings, and caps a batch at 50,000 embedding inputs.
Cache repeated queries. Anthropic's prompt caching prices a 5-minute cache write at 1.25x base input tokens, a 1-hour write at 2.0x, and cache reads at 0.1x or lower depending on the model. Minimum cacheable prefixes run from 512 to 4,096 tokens by model, so short retrieval contexts may not cache at all.
Re-ranking. Cohere's Rerank docs describe models that "sort text inputs by semantic relevance to a specified query," including rerank-v4.0-pro, rerank-v4.0-fast and rerank-v3.5. The page makes no benchmark claim against embedding-only retrieval.
Hybrid search. Elasticsearch's reciprocal rank fusion combines a BM25 query and a kNN query with a default rank constant of 60 and no manual weighting.
Abstention and citations. Anthropic's Citations feature is generally available and returns "the exact passages that support each claim."
The transcript's "query writing" is almost certainly "query rewriting," the standard term for reformulating a query before retrieval.
How do you measure RAG hallucinations?
The Builder says he would build "evaluation as separate components so i can measure and improve each one independently" but names no metric. Ragas Faithfulness is one off-the-shelf option. It scores a response from 0 to 1 as supported claims divided by total claims, checked against the retrieved context. That measures the "confident answers with the wrong information" problem the interviewer raises.
What is the BASWE.AI Engineer community?
The video closes with: "they have a full learning roadmap projects to put on my resume and daily calls with working ai and machine learning engineers." It asks viewers to comment "interview" for a link and gives no price.
The BASWE.AI Engineer Skool page lists $89 per month, 304 members, a 7-day money-back guarantee and a price increase at 500 members. A free tier, Become an AI Engineer, has 6.7k members and offers starter roadmaps. Bashiri Smith, the creator in the video, runs both.
One named instructor appears in the Classroom: Kitu Komya, whose MentorCruise profile reads "AI / ML Engineer & Senior Data Scientist. Shopify, UCLA, O'Reilly alum."
Which free courses does the roadmap use?
The screen recordings show lessons that link to free material:
- The "Basic Retrieval Augmented Generation (RAG)" lesson embeds "RAG from Scratch," a 2:33:11 course by Lance Martin of LangChain, published free by freeCodeCamp.
- Step 2 of that lesson links to DeepLearning.AI's "LangChain: Chat with Your Data", a free 1 hour 18 minute course taught by LangChain co-founder Harrison Chase.
- The "scikit-learn: Pipelines and Model Training" lesson links the scikit-learn MOOC, built by Inria and the scikit-learn team and published under CC-BY.
The subscription pays for sequencing, feedback and calls on top of that material.
Does the community hold daily calls?
The pitch says "daily calls." The TikTok description says "weekly coaching from ai engineers & recruiters." The Skool page says "Daily calls with AI/ML Engineers, Recruiters & Hiring Managers."
The calendar shown in the video covers July 27 to September 6, 2026. Twelve entries in that view are titled as cancelled, nine of them in August, including four days in a row from August 17 to 20. Recurring sessions do run, such as "10am - AI/ML - Kitu" on most Tuesdays and Thursdays and an interview-prep block most weekdays. The screenshot does not show calls running every day without cancellations. The calendar capture is dated August 26, 2026, 39 days before the video posted.
Bottom line
The Builder's RAG advice is sound and matches current vendor documentation. The $89 community behind the video builds its roadmap partly on free courses you can take directly, and its own calendar shows frequent cancelled sessions.
Sources
- Lewis et al., 2020 (arXiv 2005.11401): the original RAG paper.
- OpenAI Batch API: batch discount and limits.
- Anthropic prompt caching: cache pricing and minimum lengths.
- Anthropic Citations: passage-level citations.
- Cohere Rerank: rerank models.
- Elasticsearch RRF: hybrid search fusion.
- Ragas Faithfulness: groundedness metric.
- RAG From Scratch on freeCodeCamp: free course in the roadmap.
- LangChain: Chat with Your Data: free course in the roadmap.
- scikit-learn MOOC: free course in the roadmap.
- BASWE.AI Engineer on Skool: the paid community.
- Become an AI Engineer on Skool: the free tier.
- Kitu Komya on MentorCruise: instructor profile.
Published October 4, 2026. Based on a TikTok by @bashifuirkashi. Full video notes (markdown)