Personal website
Data scientist and incoming Carnegie Mellon MSCF student focused on machine learning, probabilistic modeling, large-scale data, and simulation-driven evaluation.
Kalash Kankaria
MSCF @ CMU | Squarepoint Fellow | Ex-Data Scientist @ Skan AI | B.Tech. CSE '24 @ IIT GN
About
I like to build models for messy systems, and financial markets are one of the messiest systems I have wanted to understand.
My interest in quantitative research started with a personal curiosity around AI-driven market signals and gradually grew into a deeper pursuit of financial theory, stochastic modeling, and machine learning for investment research.
I graduated in Computer Science and Engineering from IIT Gandhinagar and began my career as a data scientist at a process intelligence company. At Skan AI, I worked with noisy enterprise data, building machine learning systems for scale that could extract structure from unstructured behavior, text, and workflows. I like breaking down complex problems into actionable tasks to drive the most impact.
This fall, joining Carnegie Mellon’s MSCF program as a Squarepoint Fellow, I look forward to strengthening my foundations in finance. This, along with my strong background in mathematics, algorithms, machine learning, probabilistic modeling and programming positions me well for quantitative research roles.
Outside models and markets, I am shaped by competitive sports as much as by technical work. At IITGN, I captained the Aquatics Team and served as Vice-Captain of the Basketball Team while balancing academics, student activities, and research. Endurance sports, marathons, Hyrox, music, and travel continue to keep me adaptive, competitive, and curious.
I would love to connect over a coffee chat, call or a game of any sort to discuss more about my experiences or to learn something new!
I completed my B.Tech in Computer Science and Engineering from IIT Gandhinagar and will join Carnegie Mellon University's Master of Science in Computational Finance program. My technical interests include statistical learning, time-series modeling, hypothesis testing, model validation, signal extraction, NLP, and simulation-based evaluation.
Education
Master of Science in Computational Finance. Upcoming coursework includes stochastic calculus, options, investments, fixed income, and simulations for option pricing.
Squarepoint Fellow 2026-27 with a USD 20,000 scholarship. GRE Quant: 169/170.
B.Tech in Computer Science and Engineering with GPA 3.86/4. Coursework included probability, statistics, linear algebra, calculus, algorithms, machine learning, probabilistic ML, and NLP.
Dean's List in four semesters, NTSE Scholarship from the Government of India, and Excellence Scholarship in Sports and Games.
Experience
Redesigned an LLM-driven feature extraction engine with a vector database-backed RAG pipeline, reducing LLM calls by 35% through a hyperparameter-tuned simulation framework.
Integrated a multi-LLM inference layer using LiteLLM, enabling unified access to local Ollama models and external APIs through load-balanced routing and asynchronous distributed inference.
Structural Screen Clustering: Modeled latent webpage structure from HTML DOM, separating stable layout signals from instance-level noise to cluster and classify screens with 87% accuracy, enabling diverse subset sampling.
Hierarchical Activity Discovery: Modeled noisy sequential UI events into hierarchical activity representations, abstracting granular actions into composite multi-step workflows with linguistic and metadata enrichment
Text Mining: Pioneered a low-latency agentic workflow for email feature extraction and process classification, attaining 94%+ accuracy and reducing observation gaps by up to 50% for investment banking clients.
Intent Modeling: Built unsupervised semantic clustering pipeline for email action summaries, normalizing them into canonical intents and broader categories with 68% accuracy, supporting dynamic classification of new intents.
Model Benchmarking: Researched transformer architecture models for event grouping and task identification utilizing fine-tuning and transfer-learning on large datasets, averaging 94% similarity score.
Projects
Investigated selective data removal without retraining using closed-form updates and Laplace, variational inference, and Markov Chain Monte Carlo approximations for posteriors.
Achieved exact posterior recovery for linear regression after deleting up to 50% of training data, and evaluated nonlinear classification unlearning with 75-79% retraining-gap closure and 2.0-2.9x faster updates.
Benchmarked neural network architectures against XGBoost and other tree-based models on tabular datasets, identifying regimes where each model class generalized better.
Applied feature rotation, smoothing, embeddings, augmentation, and regularization; evaluated Hopular and TabPFN against tree-based baselines.
Built a local Chrome extension that extracts job details from LinkedIn and other job portals, opens an editable review form, and saves approved applications to an Excel internship tracker.
Paired the extension with a local Python writer service so workbook data stays on the user's machine and duplicate LinkedIn postings are detected before saving.
A deployed trivia application for playing quiz-style rounds in the browser.
Hosted as a live web app on Firebase.
Skills
Additional
Gold Medal for Overall Outstanding Performance in Sports, Gold Medal for Outstanding Performance in Aquatics, and Bronze Medal in Inter IIT Aquatics Meet 2023.
Interests include poker, chess, and Matiks.
Contact
Reach out by email or connect through GitHub and LinkedIn.