Candidate Information:
- Candidate Information: Md Tareq Shah Alam
- Status: Currently working as an AI/Data Engineer with professional experience in AI engineering, machine learning, LLM-based applications, and automation.
- Location: Dhaka, Bangladesh
- Interview Date: 28/10/2025
Overview of Interview Process:
Initial Screening
The interview started with a brief introduction. I was asked to introduce myself and explain my professional background and experience.
I talked about my early experience at Brain Station 23, where I started my professional career, followed by my experience at FinTech Point. I then explained my transition into AI engineering and the type of AI, machine learning, and LLM-related work I have been doing.
- Duration: 1 hour
- Method: Face-to-face
- Focus: Background, professional experience, and AI/ML knowledge
- Key Questions: Introduction and discussion about previous work experience
- Obstacles: No particular obstacle during this stage
Technical Round
The technical round covered a wide range of topics, from fundamental machine learning concepts to deep learning, transformers, LLMs, and LLM application frameworks.
- Duration: 1 hour
- Method: Face-to-face
- Focus: Machine Learning, Deep Learning, NLP, LLMs, Transformers, LangChain, and LangGraph
- Key Questions: The interviewer asked questions on RNN, LSTM, ConvLSTM, precision, recall, F1-score, cross-validation, data handling, bias and variance, Transformer architecture, attention mechanism, QKV, LLM architecture, LangChain, and LangGraph.
- Obstacles: Some of the questions required understanding both the theoretical concepts and how they are applied in real-world AI systems.
Machine Learning Concepts
The interviewer asked me about several fundamental machine learning concepts. I explained precision, recall, and F1-score, including what each metric represents and when they are useful. I was also asked why cross-validation is used, and I explained how it helps evaluate a model more reliably on unseen data.
We also discussed data handling, where I explained the general steps involved in preparing and processing data before using it for model training.
RNN, LSTM and ConvLSTM
I was asked to explain RNN and how it works with sequential data. I explained the basic idea of recurrent neural networks and how they maintain information from previous time steps.
The discussion then moved to LSTM and ConvLSTM. I explained why LSTM is preferred over a basic RNN when dealing with long-term dependencies and also discussed how ConvLSTM combines convolutional operations with the LSTM architecture for handling spatial-temporal data.
Bias and Variance
I was also asked about bias and variance. I explained the concepts of underfitting and overfitting and discussed different approaches that can be used to reduce high bias and high variance in machine learning models.
Transformer Architecture
The interview then moved into Transformer architecture. I explained the main components of a Transformer and how its different layers work together.
I also explained the attention mechanism and how it allows the model to capture relationships between different tokens in a sequence.
Attention and QKV
The interviewer went deeper into the attention mechanism and asked me about Query, Key, and Value.
I explained the purpose of Q, K, and V and how they are used in the attention calculation. I also explained the scaled dot-product attention formula and the role of the $\sqrt{d_k}$ scaling factor.
LLM Architecture
I was then asked about LLM architecture. I explained how modern LLMs are built around Transformer-based architectures and described the major components involved in processing input tokens and generating the next token.
I also discussed the different layers inside an LLM and how the attention and feed-forward components work together.
LangChain and LangGraph
Finally, we discussed LangChain and LangGraph. I explained what both frameworks are used for and how they can be used to develop LLM-based applications and agentic workflows.
I also explained the difference between them, particularly how LangChain provides building blocks for LLM applications, while LangGraph is useful for designing more structured and stateful workflows.
Post-Interview Reflections:
- Company Culture Insights: The interview was primarily focused on evaluating technical knowledge and understanding of AI technologies.
- Work Environment: Not specifically discussed during the interview.
- Benefits Highlight: No specific benefits were discussed during the interview.
- Evaluator Feedback: The interview provided an opportunity to discuss both fundamental machine learning concepts and more advanced AI/LLM topics.
- Suggestions for Improvement: For anyone preparing for a similar AI Engineer interview, I would recommend having a strong understanding of ML fundamentals along with practical knowledge of RNNs, LSTMs, Transformers, attention mechanisms, LLM architecture, LangChain, and LangGraph.
Additional Information:
The interview covered a broad range of topics, starting from fundamental machine learning concepts and gradually moving into deep learning, Transformer architecture, LLMs, and modern LLM application frameworks. I was able to discuss the topics based on both my theoretical understanding and practical experience working with AI and LLM-based systems.
Closing Note:
Overall, it was a technically focused interview that covered several areas relevant to an AI Engineer role. The discussion gave me an opportunity to explain my understanding of machine learning fundamentals as well as more advanced topics such as Transformers, attention mechanisms, LLM architecture, LangChain, and LangGraph.
It was a good experience and a useful opportunity to evaluate my understanding of different areas of AI engineering.
