Quantitative microstructure & autonomous systems in silicon. Causal temporal-graph networks for cross-sectional ranking, non-equilibrium econophysics, limit-order-book simulation, FPGA/SoC reference design, and biologically grounded neural architectures. Zero tolerance for lookahead data leakage or overfitted backtests.
Publishing when a model fails is more valuable than fabricating a Sharpe 3.0 through unpurged test folds. Negative results define the real boundary of what information is predictable in financial markets.
If model latency τ exceeds the decay half-life of informational advantage, the signal is worthless before execution. Specialized, quantized, low-parameter architectures beat massive deep nets in real markets.
No global normalizations. No overlapping forward-return labels. Purging by label span σ = H + 1 and enforcing a mandatory one-session execution lag. Theoretical backtests survive production only with this discipline.
Open pre-prints across market microstructure, causal graph networks, econophysics, HFT theory, and autonomous AI agents. All evaluated with walk-forward validation & zero lookahead bias.
A 12,801-parameter hybrid causal temporal graph architecture for cross-sectional equity ranking. Combines dilated temporal convolutions (57-feature, 60-session window) with dynamic correlation-based graph aggregation. Evaluated on 56 liquid ETFs across 7 asset classes with rigorous walk-forward validation — zero lookahead bias, purged label spans, chronological normalization.
An econophysics framework that interprets portfolio fragility through non-equilibrium statistical mechanics. Defines cross-sectional entropy Ht, directional sign synchronization St, and combined Phase Pressure Ψt to detect collective market regimes. Achieves CAGR 12.86%, Sharpe 0.899, max DD −35.33%.
Comprehensive theoretical analysis of AI in HFT. Proves the Latency Dominance Theorem: a less accurate model with lower latency strictly dominates a more accurate but slower one — α(τ) = α₀e^(−λτ). Formalizes the five-layer microstructure AI stack and five falsifiable empirical hypotheses about specialized vs. general architectures in adversarial LOB environments.
A transparent, exhaustive negative-result study across 20 heterogeneous assets. Pool accuracy 51.58% (near-chance), ROC-AUC 0.5059. Brier Score 0.2864 vs 0.2500 for a constant 0.5 forecast. Publishing negative results is more valuable than fabricating a Sharpe 3.0 through unpurged test folds.
Formal theory of how open-source publication, capital crowding, and arbitrage intensity erode trading alpha. Derives the gross alpha half-life t₁/₂ = log(2)/Λ and net break-even time t* from a dynamical system. Shows why open-source research advances market efficiency rather than destroying quantitative alpha.
Framework for calibrated uncertainty estimation in tool-using LLM agents. Addresses the compound-error catastrophe — if T actions each fail with probability p, success probability falls exponentially as (1−p)^T. Introduces a four-signal uncertainty vector, formal verifier, and CVaR risk controller with five execution modes.
Production-grade trading system simulations, FPGA/SoC reference designs, biologically grounded neural organs, and sublinear sparse attention hierarchies.
Multi-branch trading-system engineering foundation for deterministic simulation, risk controls, venue connectivity, FPGA-oriented gating, and production-readiness validation
Harmonic Phase Transformer with Decision Tree Ensemble for Cross-Asset Financial Prediction
Institutional market microstructure simulator built for quants tired of 'Sharpe 4.2' pandas backtesters crying when an HFT front-runs them in 400µs
Open SystemVerilog edge-AI SoC reference design with an RV32I control core, deterministic sensor pipeline, INT8 neural accelerator, secure boot, formal interface checks, and energy-proxy monitoring
Econophysics model for cross-asset risk allocation using collective entropy and return-sign synchronization
Open research architecture for generative financial time-series modeling
Browsing github.com/ak495867 ↗ — each repository links to a dedicated detail page.
Open market data and foundation model weights. huggingface.co/akhverm ↗
Free and open collection of historical market data covering equities, cryptocurrencies, forex, and macro assets across multiple timeframes. Built to spare researchers from expensive data vendor API paywalls.
Causal generative transformer architecture for financial time-series forecasting. Pre-trained on high-frequency market dynamics and optimized for cross-sectional generalization. Open weights for academic and research use.
I'm Akhilesh Varma — an independent quantitative researcher and systems engineer. My work sits at the intersection of market microstructure, econophysics, autonomous AI agents, and FPGA/SoC hardware design. No team, no institution — just code, math, and a high tolerance for reading negative results in my own papers.
Every model I publish is evaluated with rigorous walk-forward validation, purged label spans, and zero lookahead bias. I've found that being honest about failure — BPCM's 51.58% near-chance performance — is more scientifically valuable than publishing fabricated Sharpe ratios. The BPCM paper exists precisely because negative results define the real boundary of what's predictable.
My research spans cross-sectional equity ranking (RAN), non-equilibrium econophysics (CESP), HFT microstructure theory, FPGA silicon design (ORX-1), biologically grounded AI (bib), and autonomous agent safety. 92+ public repositories, 6 SSRN pre-prints, and open Hugging Face datasets built so other researchers don't have to pay data vendors.
ORCID: 0009-0003-5602-1863 · SSRN Author ID: 12995268 · akhverm@gmail.com
Research collaborations, quantitative systems work, or to talk microstructure theory.
Direct email for research collaboration and professional inquiries.
Professional profile, experience, and career network.
92 public repos — quantitative finance, AI systems, and silicon design.
6 research pre-prints in market microstructure and econophysics.
Klint-32M model and Bedrock open market dataset (146+ downloads).
Research updates, quantitative commentary, and systems thinking.
Long-form articles on quantitative research, AI, and market theory.
Verified researcher identifier for academic citation and discovery.