Math First or Build Cool Projects? A Deep Dive into Technical Learning Paths

A Real Case of Tech Anxiety: I'm Studying Matrices While My Classmates Are Earning Likes

On Reddit, a sophomore from a lesser-known Indian college (YCCE, Nagpur) posted a story that resonated widely. He's caught in a dilemma many technical learners face: Should he stick with a "math-first" hardcore approach, or pivot toward building application projects that produce quick, visible results?

This student's foundation is quite solid. He's already completed Gilbert Strang's legendary MIT Linear Algebra course (18.06) and personally implemented a Leslie matrix population model, an SVD image compressor, and linear regression from scratch. Gilbert Strang is a legendary professor in MIT's Department of Applied Mathematics, and his 18.06 course has been studied by millions since its release on MIT OpenCourseWare in 2002. What makes this course special is its emphasis on geometric intuition over mechanical computation—building a complete framework for understanding linear transformations through concepts like column spaces, null spaces, and eigenvalue decomposition. For ML/AI practitioners, linear algebra is the mathematical bedrock for understanding PCA, SVD, neural network weight updates, and other core techniques. He's currently grinding through Harvard's probability course (Stat 110), statistics, and multivariable calculus. He openly admits he "loves the math-first approach and hates memorizing formulas" and has zero interest in web development, React, or full-stack work.

But anxiety follows close behind. His classmates use MediaPipe, OpenCV, and React to build gesture controllers and AI mini-games, earning 2000+ likes on LinkedIn and winning hackathons left and right. Meanwhile, he's "still studying matrices and eigenvalues" with no "visible projects" to show and a blank LinkedIn profile. The self-doubt born from this comparison is the emotional core of the entire post.

The Mismatched Competition Between "Visible Projects" and "Foundational Capabilities"

Social Currency ≠ Career Competitiveness

This student has keenly identified a crucial contradiction: Hackathon awards and LinkedIn likes are essentially "social currency," not direct proof of career competitiveness.

Building a gesture recognition demo with MediaPipe + OpenCV has strong visual impact and high shareability, generating massive social feedback in a short time. MediaPipe is a cross-platform multimedia processing framework open-sourced by Google in 2019, with built-in pre-trained models for gesture recognition, face detection, and pose estimation—developers can invoke them with just a few lines of code. OpenCV provides hundreds of algorithm implementations from image filtering to object detection. Combined, they enable rapid prototyping of visual interaction demos, but the core algorithms (like the CNN architecture behind hand keypoint detection or temporal tracking algorithms) remain a black box to the caller. This is the essential difference between "calling a library" and "implementing from scratch"—the former demonstrates engineering integration ability, the latter demonstrates algorithmic understanding.

The path he's chosen—hand-coding SVD image compression, implementing linear regression from scratch—while "not as flashy," touches the mathematical foundations of machine learning and operations research. The Leslie matrix uses a square matrix to represent fertility and survival rates across age groups, simulating population evolution through repeated matrix multiplication, where the dominant eigenvalue determines the long-term growth rate and the corresponding eigenvector gives the stable age distribution structure. SVD decomposes any matrix into the product of three matrices U·Σ·V^T, and retaining only the top k largest singular values can approximately reconstruct the original image with far less data. These projects may seem simple, but they require thorough understanding of matrix operations, eigenvalue theory, and numerical computation.

The problem is that the "feedback cycles" for these two types of achievements are radically different. Building cool projects provides instant feedback—you can get likes within a week. But the returns on mathematical foundations are deferred until graduate school or even years into a career. This is the root cause of the anxiety.

Goals Determine the Path

This student's goals are crystal clear: aim for IIT Bombay's IEOR (Industrial Engineering and Operations Research) or ISI's M.Stat (Master of Statistics), ultimately entering operations research, quantitative research, or research-oriented data science roles, targeting salaries of 25-40 LPA and above (approximately $30,000-$48,000 annually, which is top-tier in India).

IIT Bombay's IEOR department is India's premier program in operations research and industrial engineering, with graduates widely entering McKinsey, Goldman Sachs, Tower Research, and other top consulting and quantitative firms. ISI (Indian Statistical Institute) was founded in 1931 by statistics pioneer P.C. Mahalanobis, and its M.Stat program is considered one of Asia's most rigorous statistics master's programs, with entrance exams notorious for their mathematical difficulty and acceptance rates consistently below 5%. These two programs represent the "golden pathway" from Indian academia to quantitative research and high-end data science, with alumni networks carrying significant influence in global quantitative finance and technology research.

For these goals, the answer is actually quite clear. The entrance exams for IIT/ISI (such as GATE DA and the ISI entrance test) test hardcore mathematical abilities in probability, statistics, linear algebra, and optimization—not MediaPipe project experience. GATE DA is a specialization added in 2024, covering six major modules: probability and statistics, linear algebra, calculus and optimization, machine learning, AI fundamentals, and programming ability. It demands extremely high mathematical derivation skills, with questions styled more like math department exams than engineering application problems.

Quantitative research and operations research roles are fundamentally the commercialization of mathematical and modeling capabilities. Quantitative researchers are the core roles in hedge funds and proprietary trading firms responsible for developing trading strategies and pricing models. Typical requirements include solid foundations in probability theory and stochastic processes, optimization theory, statistical inference, and efficient programming ability. Top quantitative firms like Jane Street, Citadel, and Two Sigma typically include extensive mathematical derivation problems in their interviews, testing candidates' speed and depth of mathematical thinking under pressure—difficulty sometimes compared to competition mathematics. On this track, every eigenvalue problem he solves now is an investment toward his goal.

Practical Advice for the "Math-First" Path

1. Don't Let Social Anxiety Change Your Strategic Direction

The poster explicitly stated he "doesn't need motivational speeches, just unfiltered truth." So here's the truth: If the goal is Quant/OR/Research DS, the math-first path is not only worthwhile but is virtually the only correct path. Classmates' flashy projects serve their respective goals (perhaps application development or product roles), but they represent a mismatched competition with the hiring standards of quantitative research—there's no need to benchmark against them.

2. Make Mathematical Ability "Visible"—Don't Abandon Math

The real question isn't "should I study math" but "how can mathematical achievements be seen." There's a win-win solution: Productize and visualize existing math projects.

For example, the SVD image compressor could easily become an interactive web application where users drag a slider to see compression effects at different ranks in real-time. The Leslie matrix population model could be paired with dynamic visualization charts. This preserves mathematical depth while producing "visible results." Writing these up as technical blog posts for LinkedIn or a personal website carries far more substance than a library-calling gesture recognition demo.

3. Beware the "Pure Theory" Trap

While the math-first direction is correct, completely disconnecting from engineering practice carries risks. Modern quantitative and research DS roles increasingly value coding ability—beyond Python, NumPy, and Pandas, one should gradually master engineering standards, version control, and data processing pipelines. His plan to supplement "ML/DL fundamentals" is wise, but I'd advise not putting it off until the very end.

4. Prepare a Plan B for Exam Failure

The poster asked "what if I don't pass GATE"—this shows mature risk awareness. The realistic answer: Without an IIT/ISI degree, OR/analytics roles still exist, but the bar is higher and the salary ceiling is lower. A solid mathematical foundation combined with a demonstrable project portfolio can still open doors to data analyst and junior quantitative developer positions. The key is not betting everything on a single exam, but simultaneously building transferable practical skills during exam preparation.

Timeline and Realistic Expectations

This student has laid out a remarkably disciplined study plan:

• Probability Phase: Harvard Stat 110 + MIT Statistics 18.650
• Math Deepening Phase: Multivariable Calculus (MIT 18.02) + Operations Research (NPTEL)
• Applied Bridging Phase: Matrix Methods (Strang 18.065) + ML/DL Fundamentals
• Exam Sprint Phase: GATE DA/PI focused preparation

Stat 110, taught by Harvard Statistics professor Joe Blitzstein, is the global benchmark for introductory probability courses. The course teaches probability through "storytelling," emphasizing conditional thinking, symmetry arguments, and transformation techniques, covering everything from classical probability to Markov chains. Its companion textbook is renowned for carefully designed exercises. For students aspiring to quantitative research, the probabilistic intuition built by Stat 110—how to make optimal decisions under uncertainty—is a more enduring core competency than any framework-calling skill.

This plan's direction is right, but one reminder is needed: Delayed gratification is an inherent feature of this path. The concept of delayed gratification originates from psychologist Walter Mischel's "marshmallow experiment," with subsequent research showing that those who can resist short-term temptation generally perform better in the long run. MIT economist Erik Brynjolfsson's research further demonstrates that investment in deep knowledge exhibits a "J-curve" characteristic—initial returns are lower than surface-level skills, but long-term cumulative effects grow exponentially. In the context of rapid AI industry iteration, frameworks and tools have a half-life of only 2-3 years, while mathematical foundations have virtually permanent validity, giving the "math-first" strategy significant risk-resistance advantages over the long term.

Most people on this path don't truly feel "ahead" until graduate school or even after entering the workforce. During the long accumulation period before that, social anxiety will resurface repeatedly. Whether one can withstand this anxiety and maintain a long-term perspective is itself one of the core qualities required for careers like quantitative research.

Conclusion: Choose the Pain You Can Sustain

This post resonated widely because it reflects an eternal tension in technical learning: short-term visible achievement vs. long-term accumulated core capability.

For this student, the answer isn't ambiguous. He genuinely loves mathematics, his goals are clear, and his path aligns highly with those goals. What he needs isn't a pivot, but learning to "build a display window" for his mathematical abilities while filling in necessary engineering gaps and keeping a backup plan for worst-case scenarios.

What's truly worth guarding against has never been "moving slowly"—it's "moving in the wrong direction" or "abandoning what you're truly good at mid-course to chase someone else's race." In an era where everyone chases instant feedback, those willing to endure delayed gratification for long-term value are already ahead.