Nine projects across AI & ML, quant finance, security, and development — open-source frameworks, published simulation research, and client work with the numbers attached. Every card links to a full case study.
Filter by practice area, or open any project for the full case study — problem, approach, and measured outcome.
Open-source security testing for AI agent systems — model, tools, memory, and planning loops, not just prompts. 130+ built-in attack scenarios mapped to the OWASP Top 10 for Agentic Applications, sandboxed tool backends, a LangGraph adapter with experimental CrewAI, OpenAI Agents, and PydanticAI adapters, 25 MCP security scenarios, SARIF output for CI/CD, and EU AI Act readiness scoring.
Open-source framework for quantum-enhanced quantitative finance: 159 modules, 14 subpackages, and 11 backends from Qiskit to D-Wave. Every quantum algorithm ships alongside the best classical solver for head-to-head benchmarking — QAOA and VQE portfolio optimization, six QAE option-pricing variants, quantum VaR via HHL, and 8 error-mitigation strategies. Install with pip install qufin.
Probabilistic expiry and volume models for NIFTY derivatives, validated with Brier score and MSE. Trading volume scaled at 124% CAGR to ₹440+ crore in total.
A 23M-parameter H-MLP simulating 155 economies and 45 sectors — outperforming VAR-X, DSGE, and Transformer benchmarks by 30% over 12-quarter forecasts — plus a 25-module supply chain simulator processing 1.2M IoT signals a day: service levels at 97–98%, stock-outs down 40%, logistics cost down 11.2%, MAPE down 44%. The methodology is published in two books; the H-MLP results are published research.
RFM segmentation and XGBoost propensity models across 120k customers. Repeat purchase rate lifted to 28.5% (+11.2 pp), with incremental revenue landing within 6 weeks.
GPT-4 paired with a finance simulator to replicate CXO decision scenarios for case and interview training. Shortlist rates up 22%, case-answer scores up 48%, adopted by 120+ MBA users.
Logistic-regression hiring model trained on 4 years of talent data. On-target hires reached 83% (+15 pp), with the mis-hire rate cut by 41%.
A portable security harness for CLI coding agents, written in Rust as an 11-crate workspace. It sits between the agent and the local system, intercepting tool calls with layered checks: prompt-injection detection (80+ rules with zero-width Unicode and base64 payload checks), secret redaction (52 gitleaks-compatible patterns with entropy gating), MCP tool pinning with SHA-256 pins and rug-pull alerts, shell/path deny rules, tamper-evident audit logging, and a session-replay TUI. Supports 8 agents including Claude Code, Codex CLI, Gemini CLI, Aider, and Goose.
Publication-quality plots in three lines of Rust. A pure-Rust plotting library with 16 chart types, PNG/SVG/PDF output, and a matplotlib-familiar API — a pyplot facade on top of a Figure/Axes model. Ships 7 themes, 14 perceptual colormaps, first-class Polars and ndarray integration, a WASM/canvas backend, and an embedded font for deterministic rendering, with no GPU or system dependencies. Dual-licensed MIT/Apache-2.0.
Let's discuss what you're trying to build, automate, or secure — the first consultation is free and there's no obligation. I reply within 24 hours.