Jisu Mondal

An independent venture building industrial-grade AI and automation infrastructure. DSRT develops production-ready systems where domain knowledge and machine learning intersect — power systems, signal intelligence, scientific computing, and multi-source analysis. Every system is built from first principles, deployed end-to-end, and open-sourced.

A signal intelligence system that disaggregates total electrical consumption from a single main meter into appliance-level usage — no per-appliance sensors required. Built a hybrid 1D CNN + Transformer architecture for sequence-to-sequence disaggregation with energy conservation baked into the loss function, plus Monte Carlo Dropout for calibrated confidence intervals on every prediction.

A Graph Attention Network for real-time estimation of complete power grid state — voltage magnitude and phase angle at every bus — from noisy and potentially adversarial sensor readings. Trained on 100K scenarios across IEEE 14, 30, and 118-bus systems with physics-informed loss encoding Kirchhoff's laws, plus adversarial training to survive cyberattack scenarios that break traditional estimators.

A physics-informed neural network that replaces SPICE-based circuit simulation. Predicts circuit waveforms 1000× faster than NGSpice while preserving physical accuracy and generalizing to unseen topologies. Built around a Fourier Neural Operator — chosen because circuit responses live naturally in frequency space — with full end-to-end differentiability enabling gradient-based circuit optimization.

A multi-agent AI system that fuses technical literature, satellite imagery, regulatory filings, and economic indicators into structured intelligence assessments. Built on LangGraph with separate research, analysis, verification, and writing agents. Fine-tuned Llama 3.1 8B with QLoRA, combined with advanced RAG (hybrid retrieval + cross-encoder reranking + HyDE) and a knowledge graph for temporal reasoning. Every output is cited.

Founded with the ambition to build AI-native hardware and autonomous systems — the integration layer between intelligent software and physical infrastructure. Robotic Rocks builds foundational technology that powers next-generation industrial automation: hardware designed from the ground up for AI workloads, not retrofitted from legacy systems.

An early ambitious attempt at building a unified OS for construction enterprises — covering project management, resource allocation, financial tracking, compliance, and on-site coordination in one integrated platform. Designed the full data model spanning projects, contracts, equipment, and personnel. The execution gap exposed the engineering depth required for enterprise platforms — directly shaping my current philosophy: build the foundation deeply before the platform.

An early attempt at a unified analysis engine ingesting diverse global sources — news, economic indicators, geopolitical events, scientific developments — into integrated cross-domain assessments. Built initial multi-modal data fusion modules but the scope exceeded the engineering foundation available. WAE's vision lives on in DSRT INTEL, executed properly with narrower scope, deeper engineering, fine-tuned models, and structured citation.

Building a strong foundation in electrical engineering while parallelly pursuing advanced AI systems research. Core coursework spans the mathematical and physical principles that underpin modern intelligent infrastructure — the same principles powering DSRT's product line.

Completed my schooling. Beyond academics, this is where I began my journey into engineering and computer science — self-taught full-stack development (HTML, CSS, Django) and data science (Python, Scikit-Learn, Pandas), shipping my first real projects before college. Also represented the school in national-level olympiads and competitive painting.