WRITING & THOUGHTS
Recent Posts
Articles on AI infrastructure, data engineering systems, and applied ML. Published on LinkedIn.
RECENT POSTS
Architecture of Thoughts — my new book exploring the structure of reason, criticism, and consciousness. It dives into how thoughts form beneath awareness and how beliefs settle. This is not motivation or surface-level philosophy—it's the mechanics of thought. Now available.
SAGUS: Frontier Continuous-Time AI Infrastructure — technical dialogue and architectural discussion with Abhinav Pandey (CEO of Quantistic) regarding the conceptual design of SAGUS as a Model Control Plane (MCP). Evaluated continuous-time state manifolds, dynamic execution control, and persistent graph reasoning for enterprise AI systems.
Exploring Adaptive RAG Systems with Dynamic Routing — a deep-dive into query complexity detection and routing across retrieval strategies. Covers dense + sparse hybrid search, confidence-based fallback chains, multi-hop reasoning, and contextual re-ranking — moving beyond static RAG pipelines toward truly intelligent retrieval.
QU-SSM: Breaking the KV-Cache Bottleneck — introducing Quasi-Unitary State Space Models and Isometric Associative Memory (ISOM). Achieves O(1) constant inference memory and linear sequence scaling through continuous-time unitary state manifolds, eliminating dissipative memory decay and quadratic Transformer scaling across infinite horizons.
Realtime AWS Data Lakehouse — built an end-to-end real-time streaming data pipeline on AWS. Ingests structured and semi-structured data using Kinesis + Kafka, transforms via Apache Spark (PySpark), and lands into a scalable lakehouse architecture with fault-tolerant ingestion and schema enforcement. Designed for high-throughput analytics workloads.
Manakizharchi AI — an end-to-end AI system built to analyze emotional sentiment and audience perception across textual content and Instagram posts. Combines classical NLP with LLMs and RAG to model complex human responses like ambiguity and sarcasm. Skills: LLM, Retrieval-Augmented Generation (RAG), +1 skill
PsychTutor — an adaptive AI tutoring system for psychology education. Uses LLM-powered conversation, Socratic questioning, and prerequisite-aware concept delivery to guide learners through complex psychological concepts. Built with a retrieval-augmented backend and session-aware state management.