Inside a Real AI Engineer Interview (Mock Interview + Evaluation)

i saw agent developers also called Ai enginner :v

00:06 - Discussion on document processing needs and AI accuracy challenges. 02:18 - Candidate highlights diverse internships and experiences relevant to AI engineering. 06:23 - Classifying movie reviews using NLP techniques. 08:32 - Discussion on LSTM and transformer models for sentiment analysis. 12:28 - Designing a chatbot for querying invoices using RAG techniques. 14:32 - OCR text extraction requires advanced handling for structured data. 18:32 - Using KNN for effective prompt retrieval enhances contextual understanding. 20:16 - KNN utilizes distance metrics to find relevant document embeddings. 23:41 - Discussing methods to identify cheaters in a coding platform using ML techniques. 25:14 - Identifying cheating in coding contests through outlier detection. 28:42 - Designing a system to identify cheating sources using graph databases. 30:27 - Identifying sources of cheating is complex in AI scenarios. 33:48 - Understanding data structures and ML metrics in AI interviews. 35:29 - Precision and recall are essential metrics in evaluating model performance. 38:58 - Explains object detection using pre-trained models and performance evaluation metrics. 40:44 - Explanation of YOLO's bounding box predictions in detection tasks. 44:20 - Discussing challenges in naming AI models during a mock interview. 46:08 - AI engineers prioritize local models over cloud-based ones for privacy reasons. 49:39 - Shift towards specialized lightweight models for improving AI collaboration efficiency. 51:19 - Agent coordination in AI can improve but isn't sufficient yet. 54:49 - Discussing the limitations and improvements of a new code analysis tool. 56:28 - Open sourcing a project helps in obtaining valuable feedback through community testing. 59:42 - Interviewers prioritize logical problem-solving over perfect answers. 1:01:17 - Fine-tuning is essential for specialized domain knowledge beyond RAG capabilities. 1:04:35 - Minimizing systems enhances accuracy in AI solutions. 1:06:16 - Understanding theoretical concepts enhances practical skills in AI engineering. 1:09:23 - The interview concludes with a positive farewell.

I thought this interview was more around Generative AI. I need interview on Applied AI involving agents not just RAG. Thanks for this btw. Would love to have someone from top tech company etc. for this if its possible.

Corrections: KNN is implied as clustering algorithm while it is actually a supervised classification algorithm. If you liked this, a subscribe would mean a lot :) If you want to participate in this series, send an email to thinkinmodelsofficial with resume attached For Mock Interviews and Personal Sessions: /karthik_varma12/ Follow for More: Youtube : Instagram: /thinkinmodelslabs Twitter: /Thinkinmodel

KNN Is a supervised classification algorithm not clustering.

We can improve attention by using two separate techniques 1. memory Saving technqiues: (comparison still O(n^2) but fast due to less blocks) a) MQA --> multiple query but one key and value blocks b) GQA --> multiple query but grouped key-values blocks 2. reducing time complexity: a) random attention (attending random tokens from a current one and not all) b) sliding window (attending the prev and next tokens only within the respective window) c) there are many variants of sliding window d) Linear Attention (we do Q(Kᵀ V) instead of (QK^T)V)

pls make a video on ai system design is it same to software system design

He has no idea of what he is saying. Just throwing in random technical jargons

Please bring more such videos! Its such a great help to someone like me who is a fresher and stuggling with the job market and interviews right now :)

That's a great video..but his ans were a bit outdated.....and he is not going indepth to explain the differences...

20:40 KNN typically uses Euclidean distance to identify the nearest data points to a given query point. In contrast, cosine similarity measures the angular difference between vectors rather than the absolute distance between their endpoints. Consider two vectors that have a very small angle between them but differ significantly in magnitude—one is short while the other is long. Using Euclidean distance, the endpoints of these vectors may appear far apart, suggesting low similarity. However, from a directional perspective, the vectors are actually quite similar because they point in nearly the same direction. In such cases, relying solely on Euclidean distance can be misleading. Cosine similarity is often more effective because it focuses on orientation rather than magnitude by computing the cosine of the angle between vectors and the query vector. Of course, in practical systems, cosine similarity is not computed against every vector exhaustively. Various optimization and indexing techniques are used to efficiently retrieve the most relevant vectors.

please make more similar videos

very informative, please keep this series continued

Hi, Thank for the Mock Interview , Just one suggestion : Please in end of the video provide the question - response specifically those are incorrect response from candidates

hi sir can u bring next interview video on ai generalist

For 100 documents question can't we just first parse those details and map those details into a database using a model and then when user gives a query we generate a SQL query for it and extract the relevant data for the user query from the database?

the question was not clear more for classification of review , said 500 then it is not real time data ,so it it is simple suprvised problem , we have data set that has features then we are going to use classical mL for classification and there is parameter one versus rest for multiclass , it we have just 2 columns one is for text and other is target then need emedding but word2vec or Glove not provide contexual embedding which will might wrong classification regards

and how someone use for such limited data neural archtiecture RNN, or its varients even said transformer .... so if it is text embedd it and bring it into tabular fom use any ML algorithm for classification

Please make more videos ❤

Very good learning and exposure to how AI/ML interviews are! Thank you very much for making this available for us.