Research • Learning Sciences & Quantitative Finance
Mahdeen
Sameer Khan
Bringing the analytical rigor of investment finance to the question of how children learn to read
Bringing the analytical rigor of investment finance to the question of how children learn to read
Sameer came to education research the long way around, through a bank's credit committee. After studying Computer Science at Colby College, he spent his early career in finance: evaluating commercial loan requests as a credit and investment analyst, then moving into private equity data analysis and investment banking. The work was quantitative and consequential. Build the model, defend the assumptions, and accept that a decision would be made on your numbers.
The reason he did not stay was personal. Watching his younger brother and his sister Sarah struggle in school, despite being plainly capable, left him with a question he could not put down: why do some students absorb information effortlessly while others work twice as hard for half the result? He went deep into memory research, mnemonic systems, and spaced repetition, and arrived at a conclusion that reorganized his thinking. "Photographic memory" is not innate. Learning is not a fixed trait. It is a function of cognitive architecture, and architecture can be designed.
What struck him next was the gap between his two worlds. In finance, nobody commits forty-five million dollars against an unvalidated assumption. In education, unvalidated assumptions about who can learn get committed against children's futures every day, and the field often lacks the measurement infrastructure to catch it. That gap became his research agenda. He has not left finance behind. He is using it.
Both of his fields ask a single question: given limited resources and imperfect information, where does an investment actually pay off, and how would you know? Education has a measurement problem, because outcomes are delayed, confounded, and expensive to observe. Finance has spent a century building tools for exactly that condition.
Difference-in-differences and propensity score matching are not finance methods or education methods. They are the methods for establishing that one thing caused another when the experiment you want is the experiment you cannot run. Sameer works across both fields because they are one discipline pointed at two questions, and because the field that allocates capital most rigorously has something to teach the field that allocates opportunity.
Underwriting $45M+ in commercial loans taught him to isolate signal from noise, stress-test assumptions against scenarios that had not happened yet, and quantify uncertainty before capital was committed.
At the READS Lab he applies causal inference and machine learning to literacy interventions, asking whether they genuinely move comprehension or merely appear to.
The rigor that protects capital is the rigor that should protect a student from an intervention that does not work. Same methods, same standard of evidence, higher stakes.
Through Sameer's Network Organization and partner affiliations, he supports underserved students in Lahore, Pakistan, alongside ongoing work in Bangladesh and Maine, building learning resources for children who lack access to them.
How do children develop reading fluency and comprehension, and can that development be measured continuously rather than episodically? He builds automated oral reading assessments that capture developmental trajectories, integrating cognitive models of reading acquisition with educational measurement principles to reduce the burden on teachers without sacrificing validity.
Which interventions actually cause learning gains, and where is the marginal dollar of educational spending best deployed? Using difference-in-differences, propensity score matching, and longitudinal fixed-effects models, he evaluates returns on educational investment against the same standards of evidence applied to capital allocation.
How do interface design decisions in digital texts affect cognitive load and comprehension strategy development? He examines how learners navigate multimodal content, particularly those who rely on screen readers or operate across languages, and how Universal Design for Learning can be built into adaptive platforms rather than retrofitted onto them.
Financial capability is usually treated as a content gap. He treats it as a learning design problem: what does it take to build genuine financial reasoning in learners who have been structurally excluded from it? This line connects both of his fields directly, spanning K–12 learners and adult professionals in high-stakes financial settings.
As a Quantitative Analysis Research Assistant, Sameer develops automated reading fluency assessment systems that analyze classroom audio recordings using advanced speech processing. Working with teacher and student recordings, he fine-tunes speech-to-text models (Whisper, Google Speech-to-Text) and builds end-to-end pipelines that convert audio to text, then transform speech into vector representations. These embeddings capture prosodic features, emotional tone, sentiment patterns, and vocabulary networks across different grade levels. By analyzing how these linguistic and affective dimensions evolve developmentally, the system identifies reading fluency indicators while reducing manual coding by 60% and maintaining 90%+ accuracy. He also conducts causal inference analyses (difference-in-differences, propensity score matching) to evaluate how vocabulary and background knowledge interventions influence comprehension trajectories.
Sameer continues to practice as an investment banking analyst, building automated discounted cash flow and market comparables models that reduce a manual, error-prone workflow to a reproducible pipeline. He previously served as a Credit and Investment Analyst at Androscoggin Bank, where he evaluated $45M+ in commercial loan requests across mid-market deals of $1M–$20M, built financial models, and conducted macroeconomic and industry research informing underwriting and portfolio strategy, including a $6.5M green energy retrofit loan supported by a ten-year DSCR model. Before that, at FoW Partners, he supported data-informed investment analysis, financial modeling, and strategic forecasting as part of the data science team. These are not two careers running in sequence. They are one analytical practice applied to two domains, and each sharpens the methods he brings to the other.
The practice is not an abstraction. It is three sets of colleagues and three buildings, held together by one standard of evidence carried between them.
Learning sciences is a team discipline. The questions that matter, how children build knowledge and what evidence should count as proof, get argued out with the people in this room long before they reach a paper.
MBA coursework brought an entrepreneurial mindset to the education question. Rigorous research answers whether something works; building something that survives contact with the real world, funds itself, and actually reaches the students who need it is a different discipline. This is the room where the two stopped being separate careers.
It started with data analysis and grew into investment. He came in to work the numbers, then moved into building the models that shaped real decisions, learning along the way that a model is only as good as the assumption you are least willing to defend. Every method on this site traces back to that private equity experience, and the habit transfers directly to asking whether an educational intervention actually worked.
Sameer investigates reading fluency development through automated assessment that combines cognitive models of reading acquisition with machine learning. The system analyzes 5,000+ student recordings using speech recognition (Whisper, Google Speech-to-Text) and forced alignment to extract developmental indicators: words per minute, accuracy patterns, self-correction behaviors. This provides insight into how children progress through stages of reading development while reducing manual coding by 60%.
60% reduction
90%+ rate
5,000+ samples
Sameer applied cognitive load theory and Universal Design for Learning to develop an accessible financial literacy platform for blind and low-vision learners, scaffolding executive function and reducing extraneous cognitive load through ARIA landmarks, voice navigation, and Braille display simulations. The platform achieved 80% task completion against a 48% industry baseline, with learners showing a 55% increase in financial confidence. The domain expertise came from his finance career; the design method came from learning science. Neither alone would have produced the result.
80% vs 48%
50 learners
55% increase
Sameer uses causal inference to understand how targeted interventions support reading comprehension development across ages, employing difference-in-differences and propensity score matching to evaluate how background knowledge building and vocabulary instruction influence oral reading comprehension trajectories. Uniquely, the design spans both K–5 classrooms and adult professionals learning under pressure inside financial institutions, testing whether comprehension mechanisms hold across radically different stakes and developmental stages. He developed R Shiny dashboards that visualize individual developmental pathways, helping practitioners identify who may benefit from additional support.
DiD, PSM
R Shiny
K-5 schools, Adults (Investment Banking)
In his finance practice, Sameer builds automated discounted cash flow and market comparables models that reduce a manual, error-prone analytical workflow to a reproducible pipeline. Earlier, as a credit and investment analyst, he evaluated $45M+ in commercial loan requests across mid-market deals, including a $6.5M green energy retrofit financed on the strength of a ten-year DSCR model he built. The underlying concern, making a rigorous method fast enough that practitioners will actually use it, is the same concern driving his automated reading assessment work.
$45M+ evaluated
DCF, Comparables, DSCR
$1M – $20M
Sameer developed an AI-powered learning tool grounded in Zimmerman's self-regulated learning framework to support students with attention difficulties, integrating speech-to-text and NLP to scaffold executive functions: working memory, planning, self-monitoring. It promotes autonomy through adaptive support that gradually fades as students develop their own metacognitive strategies. The system models the cyclical nature of self-regulation (forethought, performance, self-reflection) to help students become more independent learners.
Risk modeling applied to student outcomes rather than credit outcomes. Sameer built early warning systems for academic disengagement using machine learning (random forests, XGBoost) to identify at-risk students with 91% accuracy from longitudinal patterns. The system is grounded in expectancy-value theory and stage-environment fit theory, recognizing how developmental transitions and motivational beliefs shape academic trajectories, and runs on FERPA-compliant data pipelines that support developmentally appropriate, timely intervention.
91%
RF, XGBoost
FERPA-compliant
Sameer examined how early childhood educational investments influence developmental outcomes in literacy and mathematics over a 10-year period using longitudinal fixed-effects models, framing the analysis explicitly as a return-on-investment question. Drawing on developmental science showing critical periods for skill acquisition, he identified 12% performance improvements associated with targeted PreK interventions, and developed interactive visualizations (R Shiny, Tableau) to communicate findings about sensitive periods and cumulative development to educators and policymakers.
Argues that financial literacy fails when it is treated as content to be delivered rather than as a learning experience to be designed, and maps how FinTech tools can be built into higher education curricula.
Applies adaptive learning models to financial training, personalising instruction to what a learner already knows instead of moving everyone through one fixed curriculum.
A co-authored clinical study evaluating visceral adiposity as a predictor of cardiovascular risk among patients with type 2 diabetes.
Sameer is actively seeking collaborations on research exploring how children and adolescents develop literacy skills, self-regulated learning strategies, and academic language, as well as work at the intersection of financial analytics and learning design, where he believes the most interesting problems are.
Whether you are interested in collaborative research, need expertise in automated assessment or valuation modeling, or want to discuss accessible learning design, he would like to hear from you.
Start a Conversation