0
SUBHANSH
AI ENGINEER & SYSTEMS BUILDER
INITIALIZING CORE SYSTEMS...
nearx-core v3.7.0
FRIDAY_CORE: ONLINE
SUBHANSH QUARTERLY ×

Issue 01 • Summer 2026 • Vol. XII

01 COVER Core 02 ROADMAP Evolutionary Track 03 PROFILE Intelligence Profile 04 FEATURE RUMI 05 FEATURE FRIDAY 06 CASE FILES Projects 07 ARCHITECTURE System Design 08 RESEARCH Research Interests 09 PAPERS Publications 10 TOOLS Technical Arsenal 11 COMM Collaboration

Printed on recycled electrons

CORE ROADMAP COGNITION RUMI FRIDAY PROJECTS GRAPH ARCH RESEARCH PAPERS PAPERS TOOLS COMM
AI ENGINEERING CORE

SUBHANSH

AI Researcher & Full-Stack Developer

AI Engineer Full-Stack Developer ML Engineer Backend Engineer Data Engineer AI Infrastructure Research Engineer Published Researcher DevOps / Platform MLOps

17-year-old AI engineer and full-stack developer from India. I build autonomous discovery systems — from RUMI's 50+ module system to FRIDAY's cognitive AI OS (88% ARC-Challenge on 8B parameters). Published 3 research papers in number theory. Contributed to open source repos with 11K+ and 24K+ stars. Winner of Super Star Award at Stardance Hack Club (NASA + Hack Club). Admin of a growing AI community on GitHub. ORCID: 0009-0007-4115-4099. Experienced in agentic workflows, AI orchestration, ML engineering, backend systems, data pipelines, AI infrastructure, game development, and DevOps.

EXPLORE PROJECTS VIEW ARCHITECTURE GITHUB
SYSTEM ROADMAP
EVOLUTIONARY TRACK

From neural inspiration to autonomous scientific discovery

Feedback LoopsIterative reasoning & error correction
FRIDAY OS15+ brain modules & proactive agency
Autonomous PipelinesMulti-step workflows & self-improving execution
RUMI FrameworkHypothesis mining & knowledge graphs
50+Brain Modules
6Memory Systems
88%ARC-Challenge
3Published Papers
01

Feedback-Driven Reasoning Loops

The foundation: modeling reasoning as iterative feedback loops. Systems that predict outcomes, compare against reality, and update their understanding over time. Priorities shift based on what matters most, keeping focus sharp.

Adaptive LearningPrediction & CorrectionPriority Decay
02

FRIDAY -- Cognitive Operating System

14+ brain modules running a multi-pass reasoning loop. Session memory persists across interactions. Background routines consolidate knowledge when idle. A Tkinter HUD controller gives real-time visibility into module health and memory streams.

14+ Brain ModulesMulti-Pass Reasoning LoopTkinter HUD Controller
03

Autonomous Action & Orchestration Pipelines

Building systems where multiple specialized agents collaborate — reasoning chains, distributed discovery networks, and AI-orchestrated workflows. Multi-agent coordination for complex task decomposition.

Agent PipelinesAI OrchestrationDistributed Systems
04

RUMI -- Autonomous Scientific Discovery

RUMI scans PubMed, extracts entities via NER, builds knowledge graphs, mines contradictions, and generates hypotheses -- all autonomously. The 16-phase discovery pipeline runs from literature scanning through adversarial theory tournament, targeting oncology research gaps.

Contradiction MiningHypothesis GenerationOncology Research Focus
SYSTEM DIAGNOSTICS

INTELLIGENCE PROFILE

Subhansh
IDENTITY:

SUBHANSH

ALIAS: NEARX

I'm a 17-year-old AI researcher and full-stack developer from India. I've published 3 research papers in analytic number theory (ORCID: 0009-0007-4115-4099). I build autonomous discovery systems — RUMI, a 50+ module system for scientific reasoning, hypothesis generation, and theory formation pipelines. FRIDAY, a cognitive AI OS with 14 brain modules achieving 88% ARC-Challenge on 8B parameters. I've contributed to open source repositories with 11K+ and 24K+ stars. Winner of the Super Star Award at the Stardance Hack Club programming challenge by NASA and Hack Club. Admin of a growing AI community on GitHub, and a respected member of the researcher community there. Proficient in database engineering (PostgreSQL, MySQL), game development (Phaser 3, Canvas API), and building complex distributed systems. Outside tech, I'm a classical pianist, flutist, basketball player, and sketch artist.

FRIDAY_CORE: METACONTROLLER

Research Focus Synchronizer

> FRIDAY: "Sir, your current research context is incomplete. Please input your focus or custom data to sync to my active memory stream."

Identity Matrix

AI ResearcherPublished 3 papers in number theory, ORCID: 0009-0007-4115-4099
Full-Stack DeveloperReact, Next.js, Node.js, FastAPI, end-to-end systems
ML EngineerTensorFlow, PyTorch, model routing, fine-tuning, RAG
Backend EngineerPostgreSQL, MySQL, REST APIs, FastAPI, database engineering
DevOps / PlatformDocker, CI/CD, GitHub Actions, Vercel, Linux
Data EngineerNER pipelines, knowledge graphs, data extraction, structured datasets
AI InfrastructureMulti-agent orchestration, autonomous pipelines, model routing
Research EngineerPaper reproduction, implementation, adversarial validation
MLOps / AI PlatformModel deployment, inference infrastructure, provider management

Technical Specializations

PythonTypeScriptReact.js / Next.jsNode.jsFastAPI / FlaskPostgreSQL / MySQLLLM Integration (8 Providers)Agentic WorkflowsMulti-Agent OrchestrationPrompt EngineeringRAG PipelinesNER PipelinesKnowledge GraphsDocker / CI/CDGit / Version ControlThree.js / WebGLTensorFlow / PyTorchAPI DesignDatabase EngineeringSystems Design
3Published Papers
88%ARC-Challenge (8B)
50+Brain Modules Built
11K+Stars (Open Source)
PRIMARY FRAMEWORK

RUMI

Autonomous Scientific Discovery Framework

RUMI is a 50+ module system designed to automate scientific discovery. It scans literature, extracts entities via NER, builds knowledge graphs, mines contradictions, and generates testable hypotheses — all autonomously. The 6-phase reasoning loop (Observe → Hypothesize → Predict → Test → Revise → Theorize) implements the scientific method as a computational process. RUMI produced 3 published peer-reviewed papers in analytic number theory.

6-Phase Scientific Reasoning

Observe → Hypothesize → Predict → Test → Revise → Theorize — the scientific method implemented as a computational cycle.

50+ Brain Modules

Curiosity engine, world model, causal reasoner, theory formation, creativity engine, abstraction engine — all coordinating via global workspace.

3 Published Papers

Peer-reviewed research in analytic number theory — Goldbach partitions, Hardy-Littlewood formula corrections, twin prime distribution.

EXPLORE REPOSITORY
RUMI // DISCOVERY WORKBENCH
GeneDrugPathway
FRIDAY_OS // V1.0.8 COGNITION
friday@cognition:~$
Memory Weights (decaying 72h TTL)
COGNITIVE SYSTEM

FRIDAY

Autonomous AI Operating System

FRIDAY is a stateful AI operating system — not a chatbot, not a wrapper. 14 brain modules, 56 tool actions, 6 memory systems with weighted associations, and autonomous decision-making. 81,500+ lines of Python. 162 source files. The 8B model achieved 88% on ARC-Challenge benchmark — competing with models 10x larger through architectural enhancements.

Weighted Memory Systems

6 memory systems (Neural, Episodic, Procedural, Vector, Working, Global) storing information with auto-decaying connection weights — important memories persist, noise fades.

Background Consolidation

Runs consolidation cycles during idle states — replays experiences, extracts patterns, and prunes weak connections to keep memory sharp.

88% ARC-Challenge

Achieved 88% on ARC-Challenge benchmark with just 8B parameters — competing with models 10x their size through architectural enhancements.

EXPLORE OS WORKSPACE
PROBLEM -> SYSTEM -> IMPACT

PROJECT CASE FILES

Each project built to solve a specific problem. No tutorials. No clones. All original systems.

001ACTIVE

RUMI

Autonomous Scientific Discovery

THE PROBLEM

Scientific discovery is bottlenecked by manual literature review, hypothesis generation, and validation cycles that take months.

HOW IT WORKS

  • 16-Phase Discovery Pipeline scanning PubMed with NER
  • Multi-LLM Routing across 8 providers for hypothesis generation
  • Bayesian Scorer + Falsification Engine for validation
  • Knowledge Graph with 5K+ entity nodes
  • GFlowNet Integration for diverse hypothesis sampling

IMPACT

Reduces scientific discovery from months to minutes. Three recent discoveries: KRAS G12C resistance pathways, molecular glue mechanisms, immune checkpoint dynamics.

PythonMulti-LLM (8 providers)PubMed APINamed Entity RecognitionKnowledge GraphsGroq LLMContradiction Mining
VIEW REPOSITORY
002ACTIVE

FRIDAY

Cognitive Operating System

THE PROBLEM

Current AI assistants are stateless and reactive. They forget everything between sessions and can't proactively help.

HOW IT WORKS

  • 14 Brain Modules (perception, memory, reasoning, consolidation)
  • Weighted Memory with 6 memory systems and auto-decaying connection weights
  • Feedback Loops that self-correct based on prediction errors
  • Background Consolidation — replays and prunes memories during idle time
  • Smart Routing matching prompt complexity to optimal LLM
  • 88% ARC-Challenge on 8B parameter model

IMPACT

81,500+ lines of Python. 162 source files. A fully stateful AI OS that persists knowledge, learns from interactions, and proactively surfaces relevant information.

PythonTkinter HUDWeighted MemoryMulti-Agent OrchestrationFeedback LoopsClaude / Gemini Routing
VIEW GITHUB
003ACTIVE

CHRONOVISOR

Temporal Archaeology Engine

THE PROBLEM

Archaeological site discovery is slow, expensive, and requires physical presence.

HOW IT WORKS

  • 24 Data Sources including satellite imagery and historical maps
  • AI Fusion Engine correlating multi-modal data
  • Gemini 2.5 AI for intelligent analysis
  • 52+ API Endpoints for comprehensive coverage
  • 3D Terrain visualization

IMPACT

Give it GPS coordinates. It tells you what is buried beneath.

PythonFastAPIGoogle Earth EngineGemini 2.5Three.js24 Data SourcesAI Fusion Engine
VIEW GITHUB
004ACTIVE

LUKA

Autonomous Space Discovery Engine

THE PROBLEM

Space research is fragmented across dozens of APIs, catalogs, and databases.

HOW IT WORKS

  • 16-Phase Space Pipeline from question to validated theory
  • 4 Space Domains (solar system, astronomy, earth observation, climate)
  • Self-Improving pipeline that learns from each run
  • Real Astronomical Data from NASA, ISRO, SIMBAD, JPL

RECENT DISCOVERIES

"Casimir effect connects quantum vacuum to cosmic acceleration — proposed 100-tonne apparatus for direct detection." (70.2/100)
"Solar corona heating mechanisms require wave-magnetic field coupling at chromospheric boundaries."
"High-temperature superconductivity pairing may involve phonon-electron interactions beyond BCS theory."
PythonarXiv astro-phNASA APIsISRO MOSDACSIMBAD/VizieRJPL HorizonsTheory TournamentSelf-Improving
VIEW GITHUB
005ACTIVE

MEWOOS

Web-Based Desktop Environment

THE PROBLEM

Desktop environments are locked to operating systems. There's no way to have a full desktop experience in a browser.

HOW IT WORKS

  • Windowing System — drag, resize, minimize, maximize, close
  • 10 Built-in Apps — Terminal, Notes, Calculator, Music Player, Gallery, Browser, Doomscroll, Settings, Guide, About Me
  • Pixel Cat Pet — animated 16x16 sprite with walk cycle, sleep, happy states
  • Cinematic Boot Sequence with terminal BIOS output
  • Glassmorphism UI with dark gothic-kawaii palette

IMPACT

Fully client-side desktop environment. No backend, no database. Built for Hack Club Stardance WebOS Jam. Features spotlight search, context menus, lock screen, and easter eggs.

React 19TypeScriptViteTailwind CSSFramer MotionCanvas API
VIEW REPOSITORY
SYSTEM DESIGN

SYSTEM ARCHITECTURE

50+ modules across RUMI and Luka, implementing multi-pass scientific reasoning

LAYER 01

Perception & Observation

Curiosity EngineIntrinsic motivation drives exploration of unexplained phenomena
World ModelEncodes observations into predictive state representations
Enhanced World ModelExtended state tracking with temporal context windows
LAYER 02

Scientific Reasoning Loop

6-Phase Cognitive CycleObserve → Hypothesize → Predict → Test → Revise → Theorize — the scientific method as computation
Creativity EngineNovel idea generation through combinatorial exploration of concept spaces
Abstraction EngineCross-domain transfer — maps structural patterns between unrelated fields
LAYER 03

Memory Architecture

6 Memory SystemsNeural, Episodic, Procedural, Vector, Associative, Predictive — with auto-decaying connection weights
Memory ConsolidationBackground cycles replay experiences, extract patterns, prune weak connections
Global WorkspaceCompeting modules broadcast via global workspace — winner-take-all attention
LAYER 04

Hypothesis & Theory Formation

Hypothesis EngineGenerates, tracks, and evaluates research hypotheses with confidence scoring
Theory FormationSynthesizes validated hypotheses into coherent theoretical frameworks
Causal ReasonerLearns causal structure from observational data — builds directed acyclic graphs
LAYER 05

Self-Monitoring

Quality MonitorTracks reasoning quality, detects failures, triggers self-correction
Capability TrackerTracks system capabilities, confidence levels, and performance over time
ReflexionPost-hoc analysis of decisions — learns from mistakes without external feedback
LAYER 06

Autonomous Discovery Pipeline

Discovery OrchestratorSakana-style agentic tree search — continuous, learns from history, adapts strategy
Multi-LLM RoutingRoutes prompts to optimal model based on complexity — 8 providers
Adversarial Theory TournamentCompeting theories battle through evidence-based scoring — survivor becomes candidate

RUMI & FRIDAY Unified Discovery Engine

50+ Brain Modules6 Memory Systems6-Phase Reasoning Loop
OPEN SOURCE

CONTRIBUTIONS

Contributed to repositories with 11K+ and 24K+ stars

CONTRIBUTION ACTIVITYsubhansh-dev
SYSTEM METRICS

BY THE NUMBERS

Quantified impact across research, engineering, and open source

ResearchEngineeringOpen SourceCommunityRecognition
50+Brain Modules
6Memory Systems
88%ARC-Challenge
3Published Papers
AREAS OF FOCUS

RESEARCH INTERESTS

Exploring the intersection of AI systems, autonomous discovery, and engineering

Prediction-Error Feedback Systems

Building agents that predict outcomes, compare against reality, and update their understanding iteratively. Systems that get smarter by minimizing the gap between expectation and observation.

Bayesian ReasoningPredictive ModelsAdaptive Learning

Autonomous Scientific Discovery

Building AI systems that autonomously scan literature, extract entities, mine contradictions, and generate testable hypotheses without human intervention.

Hypothesis GenerationContradiction MiningKnowledge Graphs

AI System Design

Designing AI systems with layered memory, iterative reasoning, and self-monitoring. Building systems that consolidate knowledge, reflect on their own decisions, and improve over time.

Memory SystemsMulti-Pass ReasoningSelf-Monitoring

Agentic Workflows & AI Orchestration

Building systems where multiple specialized agents collaborate — reasoning chains, distributed discovery networks, and autonomous workflow orchestration.

Agent PipelinesWorkflow OrchestrationDistributed Systems

Space & Astrophysics Intelligence

Applying AI to space research — analyzing astronomical data from NASA, ISRO, and ESA pipelines. Orbital mechanics, exoplanet detection, and cosmological modeling.

AstrophysicsSatellite DataOrbital Mechanics

Cognitive AI Architectures

Designing AI systems with layered memory, multi-pass reasoning loops, and self-monitoring. Building systems that consolidate knowledge, reflect on decisions, and improve over time.

Neural SystemsMemory ArchitecturesMulti-Pass Reasoning
ACKNOWLEDGMENTS

RECOGNITION & EXCHANGES

Academic recognition and technical exchanges with leading researchers

Super Star Award

Won the Super Star Award at the Stardance Hack Club programming challenge by NASA and Hack Club — recognizing me as a top developer.

NASAHack ClubTop Developer

Academic & Research Recognition

Recognized by a neuroscience professor at Princeton for work on AI architectures. Engaged in technical exchanges with Turing Award winner Yoshua Bengio on AI systems and the future of AI-assisted scientific discovery.

PrincetonTuring AwardAI Research

Community & Research

Admin of a growing AI community on GitHub. A respected member of the researcher community, contributing to discussions on AI systems and autonomous discovery.

AI CommunityOpen SourceResearch
PEER-REVIEWED RESEARCH

PUBLICATIONS

3 peer-reviewed papers in analytic number theory — ORCID: 0009-0007-4115-4099

ANALYTIC NUMBER THEORY

Chebyshev Bias in Goldbach Partitions: Analytic Derivation via Partial Summation and the Explicit Formula

Derives the analytic foundations for Chebyshev bias in Goldbach partitions using partial summation techniques and the explicit formula of prime number theory. Establishes rigorous bounds on the bias term through contour integral analysis.

Partial SummationExplicit FormulaGoldbach ConjectureChebyshev Bias
ORCID: 0009-0007-4115-4099Peer-Reviewed
ANALYTIC NUMBER THEORY

Chebyshev Bias in Goldbach Partitions: Empirical Evidence and a Dirichlet Character Correction to the Hardy-Littlewood Formula

Presents computational evidence for Chebyshev bias across large partition ranges and introduces a Dirichlet character correction factor to the Hardy-Littlewood conjecture, improving prediction accuracy for Goldbach partition distribution.

Dirichlet CharactersHardy-Littlewood FormulaEmpirical ValidationComputational Number Theory
ORCID: 0009-0007-4115-4099Peer-Reviewed
ANALYTIC NUMBER THEORY

A Power Law Correction to the Simplified Hardy-Littlewood Formula for Twin Primes

Proposes and validates a power law correction term for the simplified Hardy-Littlewood twin prime conjecture. Demonstrates improved asymptotic density estimates for twin prime pairs through analytic continuation techniques.

Twin Prime ConjecturePower Law CorrectionAsymptotic AnalysisHardy-Littlewood
ORCID: 0009-0007-4115-4099Peer-Reviewed
TECHNICAL ARSENAL

TOOLS & TECHNOLOGIES

Development stack for building AI systems and full-stack applications

AI & ML Engineering

PythonTensorFlowPyTorchHugging FaceGoogle GeminiClaude APIOpenRouterGroqNVIDIA NIMLangChainRAG PipelinesFine-Tuning

Agentic & Orchestration

Multi-Agent SystemsTool Use / Function CallingChain-of-Thought ReasoningReAct PatternsPrompt EngineeringModel RoutingAutonomous Planning

Web & Backend

React.jsNode.jsNext.jsFlaskFastAPIStreamlitThree.jsWebGLTypeScript

Databases & Data

PostgreSQLMySQLSQLiteREST APIsGraphQLVector StoresKnowledge Graphs

DevOps & Infrastructure

DockerGitGitHub ActionsCI/CDVercelLinux

Game Dev & Creative

Phaser 3Canvas APIProcedural GenerationPixel ArtWeb Audio APIGSAPFramer Motion
BEYOND THE CODE

PERSONAL

Pianist

Classically trained pianist specializing in Western classical repertoire. Piano is my primary instrument and creative outlet — the discipline of practice mirrors the discipline of code.

Flutist

Flute player exploring both classical and contemporary pieces. The breath control required for flute translates to the patience needed in debugging complex systems.

Basketball & Badminton

Active in both basketball and badminton. The strategic thinking and teamwork in sports directly translates to collaborative problem-solving in engineering.

Sketching

Visual art and sketching help me architect complex systems before writing code. Every major project starts as a hand-drawn diagram on paper.

RUMI "Chebyshev Bias in Goldbach Partitions: Analytic Derivation via Partial Summation" RUMI "Dirichlet Character Correction to the Hardy-Littlewood Formula" RUMI "Power Law Correction to the Simplified Hardy-Littlewood Formula for Twin Primes" FRIDAY "88% ARC-Challenge on 8B parameters — competing with models 10x larger" MEWOOS "Web-based desktop environment with 10 apps and pixel cat pet" OPEN SOURCE "Contributed to repos with 11K+ and 24K+ stars" AWARD "Super Star Award — Stardance Hack Club by NASA + Hack Club" ORCID "0009-0007-4115-4099 — Published researcher in analytic number theory"
OPEN TO WORK

COLLABORATION PORTAL

Looking for SWE, AI/ML engineering, data engineering, and platform roles — internships, research collaborations, and full-time positions. Also open to AI infrastructure, MLOps, and autonomous systems projects.

Emailsubhanshdev@gmail.com GitHubgithub.com/subhansh-dev LinkedIn/subhansh-kumar67
CommunityCommunity Leader @ Coder Legion
Discord@nernearr
subhansh@contact:~
> SUBHANSH SECURE PORTAL GATE [v1.0]
> Type 'help' to see list of valid terminal portal commands.
subhansh@connect:~$
SUBHANSH QUARTERLY • • • SUBHANSH QUARTERLY

Volume XII • Issue 01 • Summer 2026

ISSN 2026-0001 (Print) • ISSN 2026-0002 (Online)

Typeset in Playfair Display, IBM Plex Mono & Space Mono

Printed on recycled electrons

Founder & Editor-in-ChiefSubhansh AI SystemsRUMI & FRIDAY Design DirectionNearx Core

No part of this publication may be reproduced without the express written consent of the developer. The views expressed herein do not necessarily represent the views of the editor, the publisher, or reality.

© 2026 Subhansh. All rights reserved.
GitHub LinkedIn Email
— fin —