Projects

AI enablement with governance for companies in Mexico: shadow-AI diagnostics, custom agents connected to real systems via MCP, and team upskilling.

Infrastructure for rigorous investment research in LatAm: software, analytics, on-demand research and quantitative training.

Deterministic equity-research pipeline: five domain-specialist agents fan out in LangGraph, a supervisor QA gate retries once, and findings synthesize into a sectioned report with verdict, conviction, and confidence.

Cross-sectional risk-on/risk-off classifier from the realized return-vs-volatility slope across 64 country index series — segmented into 10 cuts and validated against FRED proxies.

Multi-asset portfolio optimization engine — mean-variance, risk parity, HRP, Black-Litterman, CVaR — with a Streamlit UI and CLI.

Python toolkit for Nelson-Siegel yield-curve modeling on Treasury and TIPS curves, with a single-page browser app.

Solow-Swan growth → CAPM → Gordon Growth → Monte Carlo: move the sliders, watch the justified forward P/E update in real time.

Has AI taken a toll on US jobs? Reproducible BLS labor-market pipeline feeding a 13-exhibit interactive Plotly dashboard served via Flask.

Leak-safe, out-of-sample text-analysis pipeline for FOMC communication. RoBERTa and lexicon tone measures tested against the policy surprise reach a rigorous null; ships a live FOMC Statement Tracker.

Factor-based country equity rotation across EM/DM/World: IC-driven factor selection and statistically validated backtests instead of narrative allocation.

Open-source (MIT) component library behind Kaxanuk's data stack: downloads, validates, homogenizes, and combines multi-provider equity and fundamental data with US GAAP-based tags and point-in-time discipline.

Multi-agent mortgage underwriting demo — six LangGraph agents collaborate to render APPROVED / CONDITIONAL / DENIED with risk score and decision memo. Live SSE stream on Hugging Face Spaces.

A family of vertical agents that materialise the «multi-agent systems» theme into concrete builds: equity research, LinkedIn writer, credit underwriting.

Reusable Obsidian-vault scaffold and Claude Code skill set for running a structured personal knowledge base.

Multi-agent system that discovers, evaluates, and ranks arXiv papers, then generates LinkedIn posts from top findings.

Claude Code plugin that runs the full construction cycle of an AI agent — from idea to shipped, evaluated, observable production agent — as a gated pipeline of skills.

Claude Code plugin that turns an approved PRD into an autonomous TDD execution loop: tier-tagged phase plans, deterministic escalation rules, and disk-backed state that survives context compaction.

Open-source Claude Code skills for the CFA/quant community — a thesis stress-tester and an agentic project scaffold — extracted from patterns proven in production research work.