TradeDesk Fulltime SDE General Onsite Interview Experience and Insights

Thanks to a friend on the forum for the referral. After passing the online assessment (OA), the HR initially said they would schedule a 75-minute system design (SD) + hiring manager (HM) interview, bu

Thanks to a friend on the forum for the referral. After passing the online assessment (OA), the HR initially said they would schedule a 75-minute system design (SD) + hiring manager (HM) interview, but a day later they said my resume was strong and scheduled a full virtual interview (VO) to be completed within a week. The OA interview experience is here: https://www.1point3acres.com/bbs/thread-1144431-1-1.html The VO consisted of four rounds: the usual 75-minute SD + HM interview, followed by one SD round and two coding rounds. The HR said the system design round consisted of one high-level and one low-level round, and the coding was a low-level (LC) question. However, my actual experience wasn't exactly the same as what the HR described; different interviewers might have different interpretations. The questions were all from interview experiences shared here: https://www.cnblogs.com/jaqennnnn/p/18936193. By the way, I should mention this guy who stole interview experiences from the forum to enrich his own blog. The first round was between SD+HM, with an Indian HM in the UK and an Indian shadow in Seattle: First, there was a 5-minute introduction, then they started asking questions about my resume. Basically, the HM pressed for technical details related to the team's areas, and then asked me to describe the most relevant project on my resume. (The following content requires a score of 79 or higher to view) This part lasted less than 20 minutes, then they went straight to the question: Design a YouTube system that can upload/play/search. Other user-related aspects like login and payment can be ignored. This actually involves three systems: Upload - CDN, chunking, resumable upload, transcoding Play - streaming, adaptive bitrate, CDN Search - indexing, ranking, autocomplete The Indian HM liked to ask follow-up questions, especially about the upload part, which took a lot of time. The whole process was basically like this: - 5 minutes: Clarify requirements - 15 minutes: Draw the upload architecture - Interviewer: What database do you use? - MySQL/Column Store for metadata, S3 for video - Interviewer: What is the metadata schema? - (Draw table structure...) - Interviewer: Why not use NoSQL? - (Explain...) - 15 minutes: Draw play architecture - Interviewer: How to choose a CDN? - (Explain...) - 15 minutes: Draw search - Interviewer: How to build an inverted index? - (Explain...) - Interviewer: What about the ranking algorithm? - Time's up... In the end, there wasn't much time left for questions, just a few simple ones. This round was probably what HR referred to as high-level design. Second round, SD (Search Engine Optimization), shadowed by a Chinese guy in Canada and a white guy in California: The following content requires a score higher than 79. You can already view it. After a brief self-introduction, the question came up: Design Ad Data Propagation. The interviewer provided two parts: the ad data database and the ad bidding front-end. Design how to pass the ad data to the bidding front-end. There was a lot of hidden information, requiring constant questioning and discussion with the interviewer. At first, I thought it was just about transferring hotspots or used data, so I drew a simple pull/push hybrid P diagram. Then the guy said he needed to transfer the entire DB table. I calculated it was 30GB of data; it turned out he needed something similar to Dropbox's file transfer mechanism, using a Distributed Queue + Scheduler + Workers. The third round was coding, with a Chinese woman and a white man in Seattle as shadows: The following content requires a score higher than 79 to view I briefly introduced myself and then got to the problem. It was on Codesignal, and they directly provided a C# interface. I said I would write Python and switched languages. This round didn't require running: [Coding website name - likely a reference to a specific problem or exercise] I hadn't practiced this problem before. I remember it being ignored in interviews a few years ago because it's infrequent, boring, and difficult to pass. However, I had practiced it on Coding Website, so I managed to write it on the spot. I first wrote a version that saves the entire path during BFS. The Chinese woman said it was fine and then asked me how to optimize it. Initially, I thought it was a coding style issue, since the code signaled an interface for solve(). Then the Chinese colleague pointed out the need to optimize space and time complexity. She suggested storing the entire path would be too time-consuming, and I proposed saving only the parent and using DFS or union/find to reconstruct the path. I wrote this version. Then, she mentioned further time optimization, suggesting prioritizing words closer to the result, likely requiring a pattern map or Hamming distance, which I hadn't considered on the spot. I mentioned that I knew of bidirectional BFS for time optimization, and she said this version was also possible, but due to time constraints and the complexity of the logic, I couldn't finish it. She briefly explained how to write it, leaving 5 minutes for questions. After some research, I realized that a pattern map was likely the method she wanted, reducing word processing time from O(2^6L) to O(L). I hadn't considered this on the spot because of the overhead of building new data structures and frequent calls. Testing with AI and benchmarks showed that it only becomes faster when wordListd is greater than 5000 characters. The fourth round was coding, with a white guy in Seattle: The following content requires a score higher than 79. You can already view it. He introduced himself and went straight to the problem in less than a minute, also on CodeSinal. However, this round required running the code, and there were a bunch of test cases. The setup was even worse than the online assessment (OA). Four visible cases could run, but 13 hidden cases would only run after summing, and only the number of passed cases and the score were displayed. The problem was JSON Validator. The guy said you couldn't use libraries and had to handle all the edge cases yourself. Actually, I saw this problem in an interview experience post, but it mentioned the second phase of the "no alcohol" coding challenge on a coding website, and since it only involved JavaScript, I didn't pay attention and assumed it was a front-end problem. The problem requires: Validate whether the input string is a valid JSON object (only a subset of the syntax needs to be validated). Required syntax rules: As can be seen from the examples, the following need to be supported: Example 1: Basic key-value pairs, string values {"a": "hi", "b": "Hello, how are you?"} Example 2: Nested JSON objects {"a": "hi", "b": {"c": "Hello", "d": {"e": "How are you?"}}} Example 3: Arrays (but the problem states to ignore them until the first two examples are completed) {"a": ["x", "y", {"b": "Hi"}]} The developer required the code to be written in three parts: handling only strings, adding functionality for handling objects, and adding functionality for handling arrays. I asked if whitespace and "\" needed to be considered, and the developer said no. So I started working on it on the spot, spending half an hour writing the string part. During this time, I tried to explain the method and discuss it, but the developer didn't react much and just kept a straight face. Then the interviewer asked me to run the code. After debugging for another 10 minutes, all four cases passed, including the two he added himself. He then asked if I had time to write the last two parts. I explained the method and wrote it using Psudo code, but it was too long and I ran out of time before finishing. He let me ask two questions and then ended the interview. The interviewer was generally not very communicative and had a stern face, only smiling slightly when I asked a question at the end. Finally, my impression is that this company has many flaws. I had heard they liked LeetCode hard questions, and it certainly lived up to that reputation. The HR person didn't mention that the coding would involve running the code with built-in test cases. Their question bank seems small, so those interested should focus on practicing interview questions and LeetCode hard questions. Also, the HR person revealed their senior salary: 200K base salary, 400K stock over 4 years, and a one-year clamp like LinkedIn, followed by quarterly vesting. I checked, and their senior salary is only equivalent to Google's 4.5... I suspect they used to offer too many high-level titles and now want to tighten their standards, but the market no longer values them. Also, their technology stack is mostly C#, so it's not very strong. I feel their level and salary don't match the intensity and difficulty of the interviews. The design questions were either too broad and didn't have enough time, or they contained too much hidden information requiring constant clarification. The coding questions were all very hard, requiring me to write everything out, optimize, or run it, and there wasn't enough time either. Summary of Problems: Unreasonable Time Allocation - Mismatch between Question Difficulty and Time Low Information Transparency - Hidden cases in CodeSignal, Vague Design Requirements Unrestrained Follow-up Questions - Interviewers kept digging into details, not giving you time to complete the overall task Lack of Cultural Engagement - 4 hours of pure tech, 0 minutes to learn about the team Unclear Standards - Breadth or depth? Working solution or optimal solution? Additional Content (2025-10-07 06:39 +08:00): I had the final round of interviews on Friday, and then I was notified of my failure on Monday. The experience was indeed terrible. The HR representative said before the interview that I would hear back on Monday after the interview on Friday. I was wondering why it was so fast. It's possible they had already decided to reject me and were just going through the motions, or the bar was extremely high, requiring everything to be perfect before considering an offer. Either way, it's a huge red flag. Additional content (2025-10-08 00:07 +08:00): The HR gave some feedback, and it seems I passed the last three rounds. There was no negative feedback in the coding round; they even mentioned strong problem-solving. It's probably a fake interview, and the position is reserved for a junior employee at H&M. Various signals were also obvious: the HR didn't respond to the onsite progress, only making a prep call very late, and before the interview, they confidently said they would give me a result on Monday after the interview on Friday. Finally, the rejection was sent via a scheduled, template email.