Apocaloptimism
The believers sound like fanatics. The fanatics sound like realists. Only one of them is right.
“The only honest position is to take both seriously, hold the tension without flinching, and build.”
Komal Mathur
I research AI and build things with it. Ranked 1st out of 2,684 CS students at Manipal Institute of Technology. Published at EMNLP.
My work focuses on natural language processing, AI safety, and the languages that modern AI leaves behind. I've done counter-terrorism research at Meta, won the CyberAI Cup 2024 — 1st out of 699 teams from 102 countries — and I'm currently building a few things I'll share when they're ready.
Experience
Intern — AI Products
Stage OTTTechnology Analyst
Bank of New York (BNY)Software Developer Intern
Bank of New York (BNY)- Built an Angular App Builder and CI/CD pipeline, reducing mobile app deployment time by 60%
- Shipped a VS Code Copilot extension leveraging internal knowledge bases for context-aware UI code generation
Counter-terrorism Researcher
Meta, Mythos Labs- Fine-tuned Llama 3.2 for automated de-radicalization monitoring with real-time risk-scoring dashboards
- Scoped and tested ML features with international stakeholders (Meta, OSCE)
- Attended PCVE conference in Tashkent, Uzbekistan, sponsored by Meta and OSCE
Technical Head & AI Subsystem Head
Research Society Manipal- Leading 200+ students pursuing undergraduate research in AI and NLP
- Organized foundational ML workshops on regression, loss functions, and optimization
Samsung PRISM Intern
Samsung R&D India- Built vector database adapter with unified interface across multiple backends for Smart Home events
- Reduced LLM fine-tuning requirements by 60% through retrieval-augmented generation
GenAI Intern
Info Edge, Naukri.com- Developed automated Question Generation pipeline with GPT-4o and MySQL, reducing manual intervention by 95%
- Streamlined prompt engineering processes, achieving 25% increase in model efficiency
Research
My research focuses on NLP, AI safety, and low-resource language processing. Full list on Google Scholar.
Bangla Sentiment Analysis using Weighted and Majority Voted Fine-Tuned Transformers
EMNLP 2023 · Conference on Empirical Methods in Natural Language Processing
REACTS: Robust Ensemble Approach with Comprehensive Temporal and Spectral Features for TTS Speech Detection
Under Review IEEE Access, 2024
Research Areas
Natural Language Processing · Transformers · LLM Hallucinations & Biases · Mechanistic Interpretability · RAG Systems · Explainable AI · Low-Resource Languages · AI Safety & Alignment
Education
B.Tech Computer Science Engineering
Manipal Institute of Technology
CGPA: 9.87/10 · Rank: 1/2,684 · 1st in CSE Department
High School (PCM)
Lotus Valley International School, Noida
98.6% in 12th CBSE (School Topper) · 96.6% in 10th CBSE
Awards
1st Place Globally — CyberAI Cup 2024
International Cyber Security Data-Mining Society (ICSDS), Auckland, New Zealand. Ranked 1st out of 699 teams from 102 countries. Developed novel ensemble approaches for TTS speech detection and network intrusion detection.
Director's Diamond Jubilee Award
Manipal Institute of Technology. Awarded for ranking 1st out of 2,684 students across the first three years.
Achiever's Scholarship
Rs. 1,01,800 for academic excellence, awarded in the 1st, 2nd, and 3rd year consecutively.
Projects
Selected work. More on GitHub.
LockStep
↗AI-assisted therapy dashboard that helps clinicians track patient progress, surface patterns across sessions, and generate structured insights for treatment planning.
Aldous
↗AI chatbot monitoring dashboard with user session analytics, chat record retrieval, and document management. Built for tracking conversational AI interactions at scale.
Delphi
↗A trading bot grounded in behavioral finance — models cognitive biases, market sentiment, and crowd psychology to identify mispriced assets and contrarian entry points.
The Logit
↗AI-powered newsletter that curates, summarizes, and contextualizes the latest developments in machine learning and AI research.
Zettelkasten
↗A scientific note-taking app based on the Zettelkasten method — interconnected knowledge cards that mirror how ideas actually link together.
Skills
Languages
Python, C/C++, Java, JavaScript, TypeScript, Swift
Tools & Frameworks
Git, SQL, MongoDB, MySQL, Angular, Django, RESTful APIs, Docker, Node.js
Coursework
Data Structures & Algorithms, System Design, Operating Systems, Database Systems, Computer Networks, Software Engineering
Opinions|
t = 1.00
The Infinite Research Loop: AGI?
Everyone has a date for AGI. The dates keep changing. Maybe the question is wrong.
“AGI is not a light switch. It is a dimmer. And someone has been turning it up for years.”
Read essay →The Alignment Tax and Who Pays It
Making AI safe costs something. The question nobody wants to answer is who bears that cost.