I have supervised more than thirty undergraduate thesis groups at BRAC University. The struggle is almost never the research itself. It is finding a topic worth researching. Disclosure: The visuals in this article were generated using Gemini, Google’s AI system, for educational illustration. A Note Before We Start I am a Senior Lecturer and researcher at the Department of Computer Science and Engineering at BRAC University . Over several years of thesis supervision, I have watched the same patterns repeat, and they repeat with remarkable consistency across batches, institutions, and specializations. Students pick topics based on what is trending, not what is needed. They build the same AI/ML pipeline over and over: find a dataset, run a few models, chase the highest accuracy number, and call it research. They spend four to eight months on a topic only to realize halfway through that it has no clear contribution, then restart from scratch under deadline pressure. They struggle to answer the most basic question at their defense: what is genuinely new about this work? There is also a subtler problem I see regularly. Many students force themselves into AI and machine learning because it feels like the expected path, even when their actual interests and strengths lie elsewhere in systems, security, HCI, software engineering, or theory. The result is technically competent work that the student cannot defend with conviction, because they never cared about the problem to begin with. This two-part guide gives you the structured process I wish I could hand every student on day one of their final year. It is not a list of topics and it is not an AI/ML tutorial. It is a repeatable pipeline you can follow regardless of your specialization, your interests, your supervisor’s domain, or your institution. If you follow it honestly, you will not change your topic in month five. I am calling it the SPOT Pipeline : Scope, Probe, Own, Test. Figure 1: The SPOT Pipeline gives undergraduate researchers a repeatable four-step process from problem identification to a confirmed research question. Source: Image generated by Gemini, Google’s AI system, for educational illustration. Part 1 covers the mindset reset and the first two steps: Scope and Probe. Part 2 covers Own and Test, data strategy, tools, responsible use of AI in your research workflow, and the five implementation mistakes that stall most final-year projects by month two. The Mindset Reset: Problems Before Methods The single most common mistake I see is students starting with a method rather than a problem. Table 1 — Mindset Reset Your starting point determines your outcome. Problem-first thinking is the only path to defensible research. Source: Generated by Gemini, Google’s AI system, for educational illustration. Research starts with a problem, not a solution. The right question is not “what can I build with this tool?” but “what problem exists in my context that has not been solved properly yet, and why?” This reordering sounds minor. It changes everything. Figure 2: Where you start determines where you end up. Method-first thinking produces work nobody asked for. Problem-first thinking produces research that contributes. Source: Image generated by Gemini, Google’s AI system, for educational illustration. The SPOT Pipeline at a Glance Table 2 — SPOT Pipeline The SPOT Pipeline in full. Each step has one concrete output. Do not move forward without it. Source: Generated by Gemini, Google’s AI system, for educational illustration. Each step has a concrete output. Do not move to the next step until you have produced that output. This discipline is what separates students who finish on time from those who are still revising their topic in month three. Step S: Scope the Problem The best thesis topics for Bangladeshi students come from observing Bangladesh, not from browsing international benchmark leaderboards. Your local context is a research advantage that students in the US or Europe simply do not have. Here is a broad map of domains where genuine research problems exist right now, distributed across different areas of CS and CSE: Bangladesh offers research opportunities that most global institutions simply cannot access. In financial technology , mobile financial services like bKash and Nagad generate millions of transactions daily, yet fraud detection in this context remains poorly studied because almost all existing research uses Western credit card datasets that look nothing like MFS transaction patterns. In healthcare informatics , clinical records across the country are fragmented, multilingual, and largely paper-based, creating a significant gap in health AI research that the international literature has not addressed. Urban systems present a different kind of opportunity: Dhaka’s traffic involves rickshaws, informal lane usage, and waterlogging that standard traffic models were simply not designed for, and the few papers that exist barely scratch the surface. Moving to agriculture and climate , crop disease detection, flood prediction, and yield estimation at sub-district resolution all require locally collected satellite and sensor data that global models do not have, making this a domain where student-level contributions can be genuinely meaningful. In education technology , adaptive learning tools for Bangla-medium students are almost entirely absent from the literature, despite the scale of the need. Cybersecurity in Bangladesh has its own distinct threat landscape, including SIM swap fraud and social engineering patterns that differ substantially from what international datasets capture. Finally, human-computer interaction research that addresses accessibility for low-literacy or dialect-speaking users is essentially non-existent, even though this population represents millions of real users interacting with digital services daily. Notice what this paragraph does not do. It does not prescribe a method for any of these domains. The method comes later, after you understand the problem. Exercise: Take thirty minutes. Write down five problems you have personally noticed or heard about in Bangladesh that involve data, decisions, or patterns. Do not filter yet. The goal is volume, not quality. You will filter in the next step. Figure 3: Seven research domains where Bangladesh-specific problems create genuine opportunities for original contribution. Source: Image generated by Gemini, Google’s AI system, for educational illustration. Step P: Probe the Literature Once you have a problem direction, the next common mistake is opening Google Scholar and searching for your topic directly. This returns hundreds of results with no orientation. You do not know which papers matter, which are outdated, or where the actual gaps are. Start With Surveys, Not Individual Papers A survey paper maps the entire field for you. It summarizes what has been done, which methods have been tried, what datasets exist, and what the authors identify as open problems. The open problems section of a good survey is effectively a curated list of potential thesis contributions. How to find them: Search Google Scholar for your domain plus “survey” or “systematic review.” Sort by date and prioritize the last three years. Read the conclusion and future work sections first. Table 3— Literature Resources Seven free resources for finding academic papers. Start broad, then narrow by domain. Source: Generated by Gemini, Google’s AI system, for educational illustration. A Tool I Give My Own Students: The Paper and Literature Review Tracker Rather than building a gap map from scratch, I have developed a structured tracker that I share with all thesis groups at BRAC University. It is free and you can make your own copy immediately. Access the template here: Paper and Literature Review Tracker The template has three tabs, each serving a different purpose in the literature review process: The tracker has three tabs, each serving a distinct purpose. The Paper Tracker is the core of the system: one row per paper, recording the title, link, access status, research gaps, dataset used, and structured answers to nine reading questions. The Q9 column, which captures what each paper leaves unsolved, is the most important output of the entire tracker. The CheckList tab acts as a reading guide, prompting you to extract the right information from each paper before filling in your Q1 to Q9 answers. It covers problem, data, methodology, results, and limitations. The Datasets tab is a running inventory of every dataset you encounter across all papers, with notes on source, size, domain, and license. This feeds directly into Step O of the pipeline, where you will need to know what data is realistically available to you before committing to a topic. Figure 4: The three-tab tracker turns scattered paper reading into a structured gap map. The Q9 column is the most important output of the entire tracker. Source: Image generated by Gemini, Google’s AI system, for educational illustration. The heart of the tracker is the nine structured questions in the Paper Tracker tab. These questions are designed to work across the majority of CS and CSE research, particularly empirical and applied work. However, they are not rigid rules. Theory-based research, HCI studies, qualitative work, and design research often follow very different structures and may not fit neatly into all nine questions. In those cases, use the questions that are relevant, skip the ones that are not, and add your own where the default questions miss something important about the paper you are reading. The tracker is a scaffold, not a checklist. Q1. What problem do the authors address and why is it important? Q2. What data is used, including source, size, timeframe, splits, and collection process? Q3. What features or inputs are used, and how were they selected? Q4. What methods or models are applied, and what is the overall pipeline? Q5. What baselines are used for comparison and why were they chosen? Q6. How is performance evaluated, including metrics and experimental setup? Q7. What are the key results with numbers, and how do they compare to prior work? Q8. What are the limitations and potential biases of this paper? Q9. What does this paper leave unsolved or unaddressed? For example, if you are reading an HCI paper that involves a user study with no dataset or baseline model, Q2 through Q6 may need to be reframed or replaced entirely. You might instead ask: who were the participants and how were they recruited, what was the study design, what qualitative or quantitative methods were used to analyze responses, and what did participants find most difficult or confusing. The spirit of the tracker remains the same regardless of the research type: extract what was done, how it was done, and what was left unresolved. Q9 is the most important column in the entire tracker. Every time you can write something specific in Q9, you have identified a potential research contribution. After reading five to seven papers in your domain, the Q9 column becomes your gap map. The patterns that appear repeatedly across multiple papers point toward your strongest thesis directions. The Datasets tab serves a different but equally important purpose. As you read, you will encounter datasets mentioned across different papers. Record them immediately with a short note about their size, domain, and license. By the time you have read ten papers, you will have a realistic inventory of what data is actually available to you, which feeds directly into Step O of the pipeline. One feature worth highlighting for group projects: the tracker has an “Assigned Person’s Name” column, which allows a thesis group to divide papers across members and maintain a single shared record. Each person reads their assigned papers, fills in the nine questions, and the gap map builds collectively. This is significantly more efficient than everyone reading the same papers independently. Checking Saturation Not every interesting problem is a workable thesis problem. Some areas are so saturated that student-level work cannot make a meaningful contribution. Signs your topic is saturated: Multiple papers reporting accuracy above 95% on the same benchmark dataset The dataset is publicly available, widely used, and several years old The method you are planning is already used as the baseline in recent papers Table 4— Saturation Saturated vs open areas in Bangladesh CS research right now. Source: Generated by Gemini, Google’s AI system, for educational illustration. A useful rule: if you can find your exact topic with the same dataset and the same method in three or more existing papers, you need to change at least one of those three elements to make a genuine contribution. Figure 5: Run this check before committing to any thesis topic. A saturated topic wastes months regardless of how interesting it seems. Source: Image generated by Gemini, Google’s AI system, for educational illustration. Where Part 1 Ends By the end of Step P you should have a shortlist of two to four candidate directions, each grounded in a real problem, each with at least one clearly identified gap in the Q9 column. A word on volume: five to seven papers is enough to get your initial orientation and identify candidate directions worth pursuing further. It is not enough for a thesis-level literature review. Before you commit to a final topic, aim to have read and tracked at least twenty to twenty-five papers in your domain. The more the better. A strong thesis literature review typically covers thirty or more papers, and the tracker is designed to scale with you across that entire journey, not just the early scoping phase. In Part 2 we move to Step O and Step T, where you will match these directions to your actual resources and validate them against a two-question test. We will also cover data strategy, tools that are realistically accessible to students in Bangladesh, how to use AI tools responsibly in your research and writing workflow, and the five implementation mistakes that stall most final-year projects by month two. Want to Go Further? If you are a student looking to start your research journey or get actively involved in research projects, I am happy to connect. You can find my published work on Google Scholar and ResearchGate , and reach me on LinkedIn where I write regularly about research, AI, and education. Reach out if you have questions about your topic, need feedback on a research direction, or are interested in collaborating. The best research often starts with a simple conversation. How to Find a Research Topic as a Final-Year CS/CSE Student in Bangladesh (Part 1) was originally published in Artificial Intelligence in Plain English on Medium, where people are continuing the conversation by highlighting and responding to this story.