Dhruv Agrawal
Technology × Finance
Systems
Systems Architecture
& Engineering
Designing deterministic, constraint-aware systems across embedded firmware, autonomous robotics, and distributed compute.
Embedded Systems
Cross-architecture firmware, deterministic migration, hardware-in-the-loop validation
Autonomous Robotics
Edge AI perception, multi-sensor fusion, constraint-aware decision systems
Distributed Systems
Verifiable compute, distributed orchestration, trust-minimised coordination
Scroll to explore
Architecture & Design Principles
I build systems that operate reliably under constraint — limited compute, adversarial environments, and real-time requirements. My focus is determinism, reproducibility, and structural robustness.
Deterministic Execution
- Reproducible build pipelines
- State-aware validation workflows
- Controlled deployment transitions
Failure-Oriented Engineering
- Stress-first testing
- Edge-case modeling
- Degradation-aware design
Resource-Constrained Optimisation
- Hardware-aware model deployment
- Latency-sensitive pipelines
- Memory-conscious execution
Validation & Verification
- Walk-forward validation
- Simulation-driven testing
- Pre-deployment correctness checks
Embedded Systems & Firmware
Deterministic Firmware Migration Framework
Designed a reproducible framework for cross-architecture firmware migration using semantic hardware abstraction layers that decouple application logic from platform-specific peripherals. The system automates code translation across microcontroller families, validates correctness through simulation-based testing, and produces deterministic builds with full traceability from source to binary.
Semantic hardware abstraction layers
Automated code translation across MCU families
Simulation-based validation pipeline
Register mapping & peripheral translation
Architecture-agnostic interface design
Deterministic build reproducibility
Cross-toolchain compilation
Hardware-in-the-loop verification
Focus: reducing manual intervention in platform migration while preserving functional correctness across microcontroller architectures.
Autonomous Robotics & Edge AI
AI Perception for Agricultural Robotics
Team RoboManipal — World Robotics Championship 2024
Perception & Decision Pipeline
Developed the full AI perception stack for an autonomous agricultural robot — instance segmentation for individual plant identification, active viewpoint optimisation for multi-angle canopy observation, 3D spatial tracking using fused camera and LiDAR data, and plant-level risk prediction for targeted intervention. The system runs real-time inference on edge hardware.
Segmentation
Instance-Level
Tracking
3D Spatial
Sensors
Camera + LiDAR
Deployment
Edge Hardware
Decision & Deployment
Constraint-aware path planning with feedback-based control loops. Plant-level risk prediction drives targeted intervention planning. Full system validated in international competition environments with latency profiling and runtime performance tuning on resource-constrained hardware.
Mentored 50+ team members across ML fundamentals, computer vision, and edge deployment workflows. Represented India at the World Robotics Championship 2024.
Simulation & Agent-Based Systems
Post-AGI Digital Economy Simulation
Built a large-scale agent-based simulation modeling autonomous economic actors in post-AGI tokenised economies. The framework introduces reputation currencies and tokenized attention markets as coordination primitives, then explores policy interventions for wealth distribution dynamics. Simulation results showed a 34% reduction in wealth concentration under targeted policy mechanisms.
Agent-Based
Architecture
Reputation Currencies
Coordination
Tokenized Attention
Markets
−34%
Wealth Conc.
Exploring how emergent coordination structures behave at scale — from trust-minimised exchange to autonomous resource allocation under structural uncertainty.
Get In Touch
Open to discussing research collaboration, systems design, and analytical problem-solving.
Dhruv Agrawal
Technology × Finance
Analytical problem-solver with a deep interest in technology-driven systems and capital markets. I study how complex systems behave under stress, how models fail in production, and where structured thinking creates the most leverage.
Quantitative Finance
Decoding
Market
Dynamics
Research across risk modeling, portfolio construction, and statistical inference in financial markets — grounded in analytical rigour, walk-forward validation, and an emphasis on understanding why models fail.
Statistical Arbitrage
Cross-Currency FX Strategy
IISc Podium Recognition
Dataset & Scope
Built and validated a market-neutral statistical arbitrage strategy on a 14-year, multi-currency FX dataset sampled at 5-second frequency. The approach isolates residual mispricings after removing dominant USD exposure, then applies systematic mean-reversion rules across currency pairs.
Gross Sharpe
3.81
Net Sharpe (1bp)
1.99
Max Drawdown
−13.6%
Profitable
12/14 Years
Methodology
All evaluation followed strict walk-forward, out-of-sample protocols — no look-ahead bias, no in-sample optimisation leaked into results. Factor exposures were decomposed to understand systematic vs. idiosyncratic return components. Every finding stress-tested against transaction cost assumptions, regime shifts, and liquidity constraints.
Max drawdown −13.6%, profitable in 12 of 14 years. Validated via walk-forward backtesting with transaction cost sensitivity across multiple cost regimes.
Risk Modeling & Early Warning
Hybrid Crash Forecasting
Developed a multi-signal crash forecasting system for Bitcoin, fusing EGARCH-based volatility regime detection, LPPL bubble diagnostics, sentiment-derived features, and LSTM sequence modeling into a unified early-warning framework. The system generates probabilistic crash alerts with a 26-day average lead time, validated through rolling-window cross-validation with strict temporal separation.
0.97
ROC AUC
89%
Precision
26 Days
Lead Time
Rolling-Window CV
Validation
Focus: identifying structural fragility in price dynamics rather than fitting to historical crash patterns. Each signal component captures a distinct market regime dimension — volatility clustering, speculative acceleration, crowd sentiment, and sequential dependencies.
Portfolio Construction
Cross-Asset Portfolio Optimization
Built a regime-aware, cross-asset allocation framework combining Hierarchical Risk Parity with dynamic regime-switching filters. The system rebalances allocations based on detected market regimes — risk-on, risk-off, and transitional — validated through strict walk-forward protocols on multi-year data spanning equities, fixed income, and commodities.
148% vs 97.5%
Return
0.41 → 0.68
Sharpe
HRP + Regime Switch
Method
Walk-Forward OOS
Validation
Designed for robustness over headline returns — emphasis on drawdown control, regime resilience, and avoiding over-fit to benign market conditions.
Contact
Connect
Open to discussing quantitative research, market analysis, and analytical problem-solving.