Ekjot Singh
Engineering low-level systems and
scalable AI infrastructure.
Founder & CEO at Metanthropic Lab and CTO at TealBase. My work focuses on high-performance inference engines, open-source BaaS architectures, and deterministic reasoning models.
EXPERIENCE
01TealBase
Engineering an open-source Firebase alternative utilizing PostgreSQL. Architecting the core database engine, real-time WebSocket broadcasting (realtime-js), and secure JWT-based auth systems with Row Level Security. Managing open-source community contributions and infrastructure scaling.
Metanthropic Lab
Architecting a frontier research institution dedicated to building verifiable, interpretable, and safe Artificial General Intelligence. Leading development on cross-model generalization methodologies like Linear Gradient Matching for dataset distillation.
PROJECTS
02Wibes Check
Owner • Public
A comprehensive web intelligence engine designed for Open Source Intelligence gathering and website security analysis.
Chromabase
Owner • Private
A powerful CLI utility that encodes massive datasets and files into resilient video formats for infinite cloud storage bypass.
Publications
05SPECIFICATION: Metanthropic Neural Ablation via Attention Refraction (M-NAAR)
Introduces M-NAAR to resolve the 'Unlearning Trilemma.' By refracting attention away from high-entropy tokens rather than destroying weights, we achieve 0.00 hallucination rates and robust deletion without lobotomizing the model.
Specification for Latent Logic Topology & Soundness-Aware Calibration
Operationalizes LLMs as engines of 'Latent Causal Chains' to solve the RLVR Convergence Paradox. Introduces the Soundness-Aware Level (SAL), a microscopic metric that predicts post-alignment reasoning performance with 87% accuracy.
The Kinetic-Potential Information Disentanglement Protocol (KP-IDP)
Invalidates the dangerous conflation that Decodability equals Causality. Introduces KP-IDP to distinguish between 'Dark Computation' (Kinetic) and 'Phantom Readouts' (Potential), solving the intervention-reversal paradox.
Module 003-CFG: Chronometric Flux Gating
A dynamic regularization protocol that eliminates Latent Manifold Collapse in Sparse Autoencoders. By treating feature importance as a temporal trajectory, CFG reduces feature absorption by 95% compared to Top-K baselines.
Technical Blog
04A collection of technical write-ups, architecture decisions, and research notes focused on database internals, systems programming, and reasoning models.
View All Entries →Solving the Unlearning Trilemma with M-NAAR
Feb 12, 2026Achieving 0.00 hallucination rates in model unlearning by refracting attention away from high-entropy tokens.
Latent Logic Topology & SAL Calibration
Feb 10, 2026Operationalizing LLMs as engines of 'Latent Causal Chains' to solve the RLVR Convergence Paradox with 87% accuracy.
Chromabase: Filesystems into Pixels
Feb 08, 2026A powerful Rust CLI utility for high-density cloud storage bypass via video archives.