CMSC 435 PROBLEM STATEMENTS
Last edited 2026-09-20. Projects released for Fall 2026, with contributed text gratefully received from our clients.
Timelines for the semester will be as called out in a separate document. How a team establishes its intermediate goals in order to meet requirements and hit hard deadlines is up to the team. I suggest that you conduct a risk assessment right away, but how long you put off discovering thorny issues is also up to the team. You can even treat the project as a big hackathon at the end if you like; it's your career, after all, and it isn't like campus doesn't promote such things. But, I bet you'll also find that hackathons are better for campus than for you, and in any event this approach has yet to end well in a 435 project. Your call.
All students on a project are equal stakeholders in the effort. No one person must work at the direction of another; we cooperate in order to win. The incentive to hold others accountable is clear: all of these projects are scoped with the expectation we'll have full effort from everyone, and there is no partial credit for partial success. We tolerate others' lack of engagement at a cost paid in our own time and grades.
The incentive to fully participate should be clear. First, the final exam is constructed to reward those who have done the work all along. It chiefly address deep technical issues involving this project. Those who didn't do the work won't know how to answer questions and pass the class. Similarly if we have done the work, but not established a paper trail to this effect along the way, then we won't be able to support answers to our questions which require reflection. There is only one snapshot in time from which to work - the end of project. Without the paper trail to analyze what the perspectives were historically, then there is not much credibility in the answers. Said another way, nobody has yet been able to recreate a project record that is credible enough for exam answers, and of course there is no ability later to go back in time to insert material into the record. (Translation: Waiting until near the final exam in order to manufacture history for convenience of answering questions is deemed "not credible".)
Second, the cover sheet submitted (as our academic integrity and intellectual property statement) lists who gets credit; no name, no credit. The decision of who signs the sheet is ultimately a team consensus. Basically the rest of the team can vote someone off the island, though this is not the common occurrence, and we'd like to have exhausted our inventory of practices to promote positive engagement before it reaches that point.
My advice: do the work and document it to pass the class. It might just be that these practices actually work too. Bonus!
I offer these projects as an opportunity for us to practice substantive application of software engineering principles. We will learn by trying them out, making design decisions and then studying the nuanced consequences. We can't close that loop if we don't have a detailed record of the decisions we made, however, and that is the most important reason for our serious obligations to log activity and articulate our reasoning as we go. Working code alone won't tell us we reached our learning objectives. Please take this process seriously from the start and we will win best value from 435.
-- Jim Purtilo
1. ADAPTIVE AI
The Digital Engagement Lab at University of Maryland uses Adaptive AI (in contrast to Generative AI) to address one overarching challenge: How can AI adapt credible information to the person and situation without changing the underlying facts? Understanding this space requires answers to a variety of important, related questions, which can only be addressed experimentally. And that is where 435 comes in.
A number of tech tasks inside these experiments are common to one another. Data ingest, prompt editing, model selection, conditioning of output to potential client agents and more are activities which are mostly independent of the scholarly questions under consideration. There is no reason to implement these repeatedly or in ways that (because of inconsistent treatment) introduce threats to validity of potential results. The 435 opportunity is to architect a framework in which all of these experiments can be conducted economically, which is to say, we can accelerate the pace of science. Scholars working in this lab should be able to 'easily ' craft a module for each new experiment, keeping their focus on specifics of the question at hand instead of needing to pay down extensive tech debt time and again.
What architecture best supports such experimentation? We don't know, but the team should interact with lab participants and try a variety of pilot exercises to assess potential ways forward. The product will be a framework of substance to sustain the lab effort moving forward. (Said another way: there is no one and done solution here.) We will recognize success when the lab participants are able to conceive of new scholarly questions, craft a plug in economically and successfully conduct new experiments on their own. If in the end it is easier (or more desirable) for the lab to craft stand alone experiments than run in our framework, then we will seriously question whether the project is a success. Along the way we need to not only handle 'right answer' checks but also usability and sustainability checks. Because content will deal with intellectual property, we need to reach success in ways that respect others' property and privacy rights. Our deliverable would be the framework, documentation and a sufficient number of specific modules to ensure our claim of success is credible. (All of the examples would be nice.)
Example experiments we know of so far are:
Our client in this work is Dr. Ronald Yaros in the Digital Engagement Lab. Team members working on this project need strong capacity to embrace ambiguity and work through the research process, but the opportunity is in pushing a powerful open source package that will see use in the community, plus potential for publication based on this work.
2. VARM2
The Neuromotor Control and Learning (NMCL) lab at the University of Maryland studies, among other topics, how people interact with robotic systems that share control of physical actions. Of particular interest is a user’s sense of agency—the experience of controlling one’s actions and their outcomes—and how that experience changes as an assistive robot assumes more control.
A previous CMSC 435 team developed VARM, a Unity-based virtual research platform in which participants use a joystick to control a simulated robotic arm while sharing control with an autonomous controller. VARM Version 1 established the basic shared-control, experimental, behavioral-logging, and EEG-synchronization capabilities needed to begin studying these questions, and has since been used for pilot research.
Our opportunity this semester is to create a fundamentally new experimental capability using that foundation. Rather than Version 1’s relatively simple reach-and-grasp task, in which a participant moves toward one object and a successful grasp ends the trial, the new system will support an entirely new timed, repeated object-manipulation task modeled on organizing pills into daily pillboxes. Participants will repeatedly use the robotic arm to grasp simulated pills from a source container, transport them through a three-dimensional environment, and precisely place them into designated day-labeled containers while operating under varying levels of shared robotic assistance and task difficulty.
The envisioned system will turn VARM from a single-action research prototype into a much richer experimental platform. It will need to support repeated grasping and precision placement, realistic object and collision behavior (including obstacle-aware path planning), failure and recovery sequences, goal-directed robotic assistance that respects the physical environment, participant familiarization and training, configurable experimental conditions and surveys, and detailed behavioral and EEG-synchronization data suitable for scientific analysis. Existing Version 1 capabilities will be reused where useful, but the new task and expanded functionality will require substantial new design.
Our client is the NMCL lab, with Geoffrey Short as the primary point of contact under the supervision of Dr. Rodolphe Gentili.
3. LEXICOGRAPHY
AI-boosted search functionality exploiting multi-language parallel data sources
The problem we address is that of limitations in the search capability at a cutting edge lexicography service specializing in PRC document tracking. Our opportunity is to leverage AI techniques to offer sharper and more responsive content to users trying to wade through what could be a tremendously large data space.
When Echtralex first started producing a marketable product that had customers, it was 2021. The item in question was a dictionary - of Chinese terms into English - published monthly in the form of a PDF that was downloadable from our website.
Even the first version we produced would have been cost-prohibitive to print, so it was and remains a digital-only product. The product served us well for many years and has a healthy following even now. But an analysis our of web traffic showed us something surprising - it was our parallel document archive (and not the book) that was getting the most attention.
This archive contains several thousand documents - all PAI from official PRC government ministries - that we originally began collecting as source material for the book. Our website offers some limited functionality to search both dictionary and the document archive, including ways for the search results to be resorted, framed, and eliminated.
But it has become clear that there needs to be a more meaningful approach to searching this data, and have the results presented in way that is both elegant, and that takes advantage if the incredible wealth of data and analysis the we have amassed and written.
One case of how the website can be improved is visible in this example:
A search of the dictionary for the term "中巴" returns the following:
Search string: https://echtralex.com/dictionary/中巴
Result: 中巴关系1 zhōng bā guān xì China-Pakistan relations 巴基斯坦总理夏巴 [snip] 中巴关系2 zhōng bā guān xì China-Barbados relations 中方对莫特利总理 [snip] 中巴关系3 zhōng bā guān xì China-Brazil relations 中国和巴西分别是 [snip] 中巴关系4 zhōng bā guān xì China-Bahamas relations 中方对戴维斯总理 [snip] 中巴关系5 zhōng bā guān xì China-Panama relations 中方注意到巴拿马 [snip] 中巴关系6 zhōng bā guān xì China-Palestine relations 阿巴斯总统是中国 [snip]
You can clearly see that the phrase "中巴关系" can be defined six different ways. The dictionary then presents multiple examples for each variant.
Search string: https://echtralex.com/doc-search/中巴
However, a search for the same term in the document archive, while producing a list of hits that can be altered (as already described), presents that data in a way that does not filter by definition. That needs to change. (The particulars of the approach we recommend can be discussed in much greater depth at a later date. We believe that there is an elegant, fairly techie solution.)
A second example with similar shortcomings can be seen in a search for the team "四点建议":
https://echtralex.com/dictionary/四点建议
四点建议 1 sì diǎn jiàn yì Four-Point Proposal (put forth by Wang Yi [snip] 四点建议 2 sì diǎn jiàn yì Four-Point Proposal (put forth by Wang Yi [snip] 四点建议 3 sì diǎn jiàn yì Four-Point Proposal (put forth by Wang Yi [snip] 四点建议 4 sì diǎn jiàn yì Four-Point Proposal (put forth by Wang Yi [snip] 四点建议 5 sì diǎn jiàn yì Four-Point Proposal (put forth by Xi Jinping [snip] 四点建议 6 sì diǎn jiàn yì Four-Point Proposal (put forth by Wang Yi [snip] 四点建议 7 sì diǎn jiàn yì Four-Point Proposal (put forth at the fourth [snip] 四点建议 8 sì diǎn jiàn yì Four-Point Proposal (put forth by Wang Yi [snip] 四点建议 9 sì diǎn jiàn yì Four-Point Proposal (put forth by Foreign Minister [snip] 四点建议 10 sì diǎn jiàn yì Four-Point Proposal (put forth by Wang Yi on [snip] 四点建议 11 sì diǎn jiàn yì Four-Point Proposal (put forth by Wang Yi on [snip]
As before, there are eleven different definitions of the term, none of which are discernible or scrutable with the current query system.
Search string: https://echtralex.com/doc-search/四点建议
There are a number of other improvements we would like to make - those that would enhance English-language searches, for instance, as well as a means of doing natural-language queries of the document archive.
The project approach for this project will be agile, and the task list progressively refined as the modeling of search and workflow firms up. Anticipate the need to conduct experiments to discover what are viable solution approaches; there is no one-and-done solution here. The deliverable must address practical needs of compatibility with existing infrastructure, maintainability and sustainability in hands of the client.
Our client in project is Echtralex, for which the point of contact is Michael Horlick.
4. NEURAL
Visualizing Electromagnetic Field Interactions and Their Effects in Neurons and Neural Populations
The process of synaptic transmission in neurons is relatively well understood when compared to non-synaptic ('ephaptic') transmission. The latter is a newer finding in the field of neuroscience, and it is challenging to understand since its mechanism cannot be easily observed. Therefore, the opportunity here is to develop a visualization and modeling tool which allows users to visualize the current experimental findings regarding ephaptic transmission (as described in research articles). The purpose of the tool is to validate existing findings, easily understand ephaptic transmission, and test various theories of its function.
Although the "right" visualization is naturally subjective and will require testing, there will be a clear emphasis on consistency with existing scientific findings and ability to reproduce previous results, as well as test new theories. Key aspects of the software that will facilitate this outcome include proper adjustment for the scale of the simulation, considerations of externally applied electric current, cell types and oscillation mechanisms, the cell/population mathematical models being used, the emergent properties being studied, the underlying topology and geometry of the network, and other factors. It is these considerations being merged together into a single, effective software that is the true challenge.
A successful project is one that demonstrates a clear alignment with experimental results and observations. This includes, but is not limited to: Empirical electrical signal/field recordings at low (individual cell - membrane potential) and high (cell networks - local field potential) levels, previous illustrations and visual observations (e.g., spatial structure of neurons), high-level empirical findings (e.g., propagation of signal activity independent of synapses via the electric field), and emergent phenomena (e.g., synchronization of adjacent cortical columns). A successful outcome is also assessed by ease of use of the tool, attention to detail (e.g., diligence with numerical precision), quality of the visualizations, and extensive testing to remove bugs/crashes/errors.
The end product will be one that can serve effectively in both an educational and research context. To precisely characterize whether the product solves the problem in these two contexts, the following questions can be evaluated: (For the educational aspect) If we showed the software to someone who knew only the very basics of neurons, could they understand and explain the principles of electromagnetic interactions quickly? (For the research aspect) Is it possible to easily and quickly test various theories or models in the software and record observations, and do these align with existing knowledge?
Our client in this project is PhD student Arya Teymourlouei in the CS department.
Resources on some of the concepts concerning this project may be found here:
5. GRADEFLOW
We are looking for a system that will streamline the process of grading exams in very large classes. Our win is in saving time while raising the grading quality and consistency.
The obvious way to grade exams would be for an instructor to write a grading key, and for the instructor and TAs to read and apply it while grading exams on paper. For a small course, with few graders (TAs) this is sufficient. However, several problems arise when trying to grade exams manually in a large course with many graders:
A system that partially automates the grading workflow would ameliorate these problems. An instructor could begin with a more concise, skeletal exam grading key, which the graders could dynamically add to and enhance on the fly (possibly with instructor approval), as unexpected situations are found during grading.
There is a commercial system that automates exam grading, Gradescope, which is available at UMD. However, it has some features that are unnecessary- even undesirable- for our purposes, and it doesn't support some other abilities that could be very helpful in teaching.
We would like to start with scanned PDFs of student exams, which would have to be uploaded to a system and somehow split up or assigned to students. Then as graders grade, they would be presented with the PDF of the exam or question being graded, for remaining exams that had not yet been graded (or where that question had not yet been graded). Graders would make deductions and comments on exams/exam questions that would be saved in the system, until grading was done, then the total score for each student could be computed. (Note that some students miss an exam for a valid reason and may not have taken it before most of the exams are being graded, so we would have to be able to add some exams later, and have graders grade them as well.)
Note that for some exams, graders may be divided up by question (three graders are grading question #1 on exams, four are grading question #2, etc.), while for short or simple assignments (perhaps some quizzes), each TA grades an entire assignment for each student. Thus, the new system must offer flexible configurations to adapt to the exams and scale of grading efforts.
Our client in this project is Larry Herman, who indeed needs more support for grading at scale. The expectation is for a reliable and highly polished U/X, protection of PII, compliance with accessibility obligations and no small measure of sustainability properties. This is not a research project - it is a hard core dev opportunity.
Larry offers some further thoughts, in no particular order, which we capture here:
6. ECOWELLNESS
Our opportunity is to highlight Nature Positive UMD via construction and deployment of an app to suggest tailored walking routes which promote biodiversity across campus.
Our goals are many. One important goal is wellness - students, faculty, staff and visitors may well benefit from a little better connection with the ecology around us on a regular basis. Another is science - if there is value in such experiences then we'd like to know how to measure it. And don't forget the education benefit of better awareness of the available ecospace.
We already have 435 experiences in this space from the Roots and Routes project, which was a successful effort with the College of AGNR. It served several educational missions and relied on the campus arboretum databases. Our present project shares many of these goals. What inspires a new build is our speculation that an app with lighter tech footprint might be more responsive to specific ecowellness needs. Instead of trying to do many things (and building ever-bigger tools) in this project we'd like to get lean and focus on wellness routes. Let's find how nimble we might be.
This project will need serious assessment of existing map options together with extensive modeling of client needs. Any user experience that appears ponderous or klonky is a fail. This is one where U/X is paramount. There are many ways to get there, whether via refactor of existing map assets or a blend with new front end designs. Which is best? We don't know but a successful team will inform us. Further, this is not some generic "go make a program" assignment where we can declare victory just by tossing a program over the wall. We will need to see successful use in deployment, which means modeling of U/X, assessment in the field with deployment and of course it is fueled by data. (An app without useful data could theoretically work but could only theoretically gets a passing grade.) We will need to work out how in the end we recognize success, which includes sustainability criteria too.
Our client in this project is Dr. Juanita Choo.
7. EXCEED
TerpsEXCEED at the University of Maryland is a unique program for students with intellectual disabilities to fully experience college with courses, residential life, campus life, and work experiences. TerpsEXCEED students are supported by a large network of trained and certified peer mentors who are UMD undergraduate students who provide academic, social, and career support to facilitate inclusive living, learning, and earning.
The key to making this happen is a shared schedule for signing up to provide mentoring support. Mentors select hours weekly that work with their own schedules. Mentor supports may vary week to week. The current system relies on shared Google calendars and sheets to handle scheduling needs, and this is the pain point: this simply doesn't scale to the size of our TerpsEXCEED system. Users spend a lot of time juggling dates and searching irrelevant data to get to their own schedule. It is confusing and error prone.
Our opportunity is to replace the scheduling system with a bespoke tool that is sensitive to the TerpsEXCEED needs. The quality of user experience is of paramount importance in this project. We anticipate the need for a fair amount of modeling of the work flow to ensure we measurably reduce time demands on participants while improving clarity. It should be "easy" (for definitions of easy we work out) for participants to find open slots and manage their time, just as it should be easy for the students to interact with the schedules and have confidence their needs will be covered. There will be several staff roles to factor in (editing any courses/content/time frames that are student specific - because courses are dropped, people change their minds, mentors change; tracking of coverage, sending announcements; individual TerpsEXCEED schedule visibility and full group visibility).
We will interpolate important requirements based on accessibility and privacy as well. We should design for 'mobile first' but do not anticipate building any native mobile apps - webapps will be the goal instead. The system will need to be resilient and sustainable after we hand off to the TerpsEXCEED staff.
Our client is Meredith Gramlich, Senior Faculty Specialist with the Center for Transition and Career Innovation in our College of Education and Associate Director of TerpsEXCEED. She will serve as our point of contact in working with the mentors and participants as we pilot potential ways forward and run experiments. TerpsEXCEED is excited to partner on this opportunity to expand inclusion on campus.