SHIVAM PRAJAPATI

Independent AI researcher and full-stack software developer with a BSc in Computer Science. Developed the Neuro-Symbolic Cognitive Kernel, integrating advanced AI methodologies, while concurrently designing a production-grade algorithmic trading system. Proficient in multiple programming languages, delivering complex systems from architecture to implementation.

• Gujarati
• Hindi
• English, Full Professional
• Gujarati, Native / Bilingual
• Hindi, Native / Bilingual

Core Research — NSCK & Cognitive Architecture

• Engineered NSCK, an open-source neuro-symbolic AGI kernel (~40K Python + 4.8K Rust LOC, 102+ modules, 1,742 tests) unifying VSA, SNNs, Global Workspace Theory, causal reasoning, and STRIPS planning; authored 10 original research papers including one published on Zenodo (Apr 2026) achieving ρ = 0.9906, ARI = 0.723, Top-1 = 78%, and +0.400 reasoning gain over baseline.
• Designed and implemented Resonance-X (RXCA), an original VSA/HDC cognitive reasoning engine featuring a TripleStore multi-hop reasoning system (conf = 1.000), a Rectified Born Rule anti-hallucination gate, liquid-γ adaptive learning, and GC-VSA H-SSM modules; architected a four-tier deployment system scaling from Raspberry Pi (D=256–512,
• Designed RSB-3 (Resonant Sequential Binding v3), a novel 4-layer VSA text generation architecture combining Random Indexing, role-frame binding, Modular Hopfield Network retrieval, Resonator Network sharpening, and a Forward Association Memory + LSH generation loop; coupled with a thin 17M-parameter fluency bridge decoder achieving 76/76 tests passing, mean coherence 0.606, and 104.7 tokens/sec on CPU at D=5000.
• Developed four original algorithms under the qFHRR system (Quantized Fractional Holographic Resonance Representation): qFHRR substrate, Resonance Gate (retrieval), Resonator Network (factorization), and Bucket Entropy Pruning for LSH optimization — grounded in arXiv-published 2025 research.
• Invented ReSC+ (Resonance Sleep Consolidation+), a background prototype-replay and bucket-hygiene daemon for MHN memory management; one of 24+ original named algorithms developed across the NSCK research program.
• Completed full architectural design for NSCK 2.0 v1: D=16,384 hypervectors (Johnson-Lindenstrauss justified, σ=0.0039, false-positive probability ~10⁻³⁷), BitNet b1.58 2B4T as foundation model, three-tier SHVKR society (Metropolitan/Suburban/Rural) with HNSW indexing, and a forward-only transplantation bridge; designed to run on consumer hardware with no catastrophic forgetting via SoCL.
• Co-developed SHVKR v26 and SoCL (Societal-Guided Elastic Weight Consolidation), a continual learning framework that uses societal consensus signals to modulate EWC and prevent catastrophic forgetting while allowing concept drift; built full constitutional governance engine v1.1 covering 7 failure modes (entropy ceiling, bond cap, cycle prohibition via DFS, 14-day shadow merge, GDELT quarantine, macro broadcast cap, Leiden stability validation).

Quantitative Finance & ML

• Built Price Signal Prediction System V4, a 5M-parameter transformer trained on ~27M windows across 527 tickers using ZigZag pivot sequences with compositional token embeddings (direction × magnitude × duration), RoPE, and cross-timeframe context features; uses MCC as the primary evaluation metric with walk-forward validation.
• Designed and built a full Autonomous Algorithmic Trading System (shadow-mode): 487-ticker 1-min OHLCV pipeline via EODHD + yfinance, Rust-based incremental feature engine (200+ features), LightGBM + Temporal Fusion Transformer models, EDGAR/SEDAR+ scrapers with local Ollama LLM extraction, and a 550K+ article Living HyperVector world state (9,500 LHVs, 5-layer, Leiden community detection, constitutional governance v1.1).

Applied Projects & Ventures

• Architected TrustGraph, a decentralized peer-to-peer social trust platform using on-device NSCK inference, Merkle event logs, cryptographic reputation snapshots, and directional trust-based social recovery; identified 3 provisional patent candidates (VSA-based collusion detection, two-party asynchronous consensus protocol, social recovery via bilateral trust).
• Sole developer of TireTrack, a full-stack tire recycling logistics platform (Flutter + Node.js + MongoDB + OpenStreetMap/flutter_map); delivered complete 9-phase build plan, real-time WebSocket chat, Twilio SMS, FCM push notifications, EXIF-stripped photo uploads, and Cloudflare Tunnel self-hosted deployment on Ubuntu.
• Co-founded GlazeBid, a B2B SaaS targeting Canadian government procurement bid estimation for glazing contractors; priced at $99/$249/$449 tiers with NRC IRAP funding identified (up to 73% dev cost recovery via SR&ED); competitive positioning built around Canadian procurement compliance gap in US-only tools.