Vipul Bajaj
I study how people and machines reason under uncertainty: markets, models, longevity, and agentic systems. I use companies to run experiments papers alone cannot.
Quantitative researcher · Founder, IntenTrails (current experiment) · Former founder, Katalyst · Former Senior Quant, Alphagrep · IIT Kanpur · Scholar
Former Senior Quant at Alphagrep, production ML at Cohesity, IIT Kanpur. I ship code, test ideas through products, and write every Sunday.
More background
Three years on HFT in China commodity markets. Domain adaptation at BMVC 2019, audio video learning at ICASSP 2021. Side projects include SaveAtlas, changiAI, and ThreadLens. I like explicit claims, quick falsification, and code that actually ships.
Research
Problems I keep coming back to, and how I try to make progress on them.
Questions
- How can we design software that actually helps us think better, rather than just helping us find old notes faster?
- Which brutal lessons from high-frequency trading actually matter when building consumer AI products?
- When do AI models make us sharper, and when are we just letting them quietly do our thinking for us?
- What does it actually take to engineer an extra decade of high-quality life, instead of just pushing numbers around on a health chart?
- If personal agents and knowledge graphs are the future, what finally replaces the infinite scroll of the social feed?
Horizons
Long-running interests I read and talk about even when they are not in this week's build.
- Longevity Healthspan as an engineering problem: biomarkers, training, sleep, and what it would mean to age slower in ordinary daily choices. Peter Attia on healthspan →
- Agentic systems Personal agents, portable identity, knowledge that travels with you instead of living inside siloed apps. Skeptical of hype labels like web 4.0, serious about the shift underneath. Maggie Appleton on digital gardens →
Method & publications
Method
Start with a hypothesis, build something small, put it in front of users or data, then write up what failed. Products are lab instruments. The weekly essays are the public log.
Selected publications & output
- Collaborative Learning to Generate Audio-Video Jointly (IEEE ICASSP, 2021)
- Curriculum based Dropout Discriminator for Domain Adaptation (BMVC, 2019)
- Explainable ML: interpretability in high stakes models
- Google Scholar profile
Reading list
- Latency & HFT. Market microstructure and the physics of execution.
- Orwell on language & power. Rhetoric, institutions, and collective behavior.
- Mental models. Cognition, bias, and decisions under uncertainty.
Principles
Beliefs I try to keep even when the company changes shape.
Operating principles (7)
- Uncertainty is the point. Markets, models, and minds are interesting because they refuse to guarantee outcomes.
- Claims need friction. A hypothesis earns trust only after data, users, or a market pushes back.
- Companies are experiments. Startups let me run studies papers cannot. The product may pivot. The method should not.
- Shipping counts. From HFT to production ML: if it is not built and measured, the research is not done.
- Structure before scale. LLMs help when memory and synthesis exist. They hurt when prompts stand in for thinking.
- Think in decades. Health, reputation, and knowledge compound. A quarter is too short a backtest for a life or a field.
- Write what failed. The weekly posts include dead ends, not just wins.
Writing
Short weekly notes on markets, cognition, longevity, and agentic systems. A new one every Sunday.
Work & side hustles
Roles, ventures, and things I have built or shipped to test ideas in the real world.
Jobs
-
Alphagrep Securities
Senior Quant Researcher. HFT strategy research on China commodity markets, based in Singapore.
-
Cohesity
Production ML & Software Engineering.
-
Gartner
Research & Analytics.
-
MPI, Uplara, NUS
Early career roles and academic research.
Products & Side Hustles
-
IntenTrails
Current experiment. Direction may change.
-
Katalyst / myKatalyst
Personal knowledge graphs and cognition tools.
-
SaveAtlas
Searchable archive for Instagram saves with a Next.js backend.
-
changiAI
Executive assistant for workflow augmentation.
-
ThreadLens
Chat analytics across WhatsApp, Telegram, iMessage.
-
acadAI
NLP for exam generation and grading. Microsoft Code.Fun.Do.
-
Explainable ML
Interpretability and transparency in high stakes models.
More projects
-
Mapped gram
Instagram saves on a map.
-
XBRL pilot
XBRL tagging automation.
-
Math of ML
ML algorithms from scratch in Python.
-
Prosthetic Arm
Embedded robotics and control.
Career timeline
- Founder, cognition and knowledge systems (Katalyst)
- Senior Quant Researcher, Alphagrep (Singapore)
- Research Engineer, Cohesity
- IITK Convocation Award; ICASSP paper on audio video learning
- BMVC paper on domain adaptation; ACM ICPC Regional Rank 52; Deloitte Grand Prize
- Microsoft Code.Fun.Do winner (acadAI)
Contact
Research
Collaborations on ML, markets, cognition, longevity, or knowledge systems. vipulbajaj2020@iitkalumni.org
Founders & partners
Technical founders, investors, and collaborators on cognition, ML, or longevity. I enjoy long conversations on healthspan, agentic systems, and where personal tools are heading. Schedule via Calendly