02 / index

Projects

Twelve RolesEvolving

Self-Hosted Multi-Agent Research

The open-source TradingAgents graph running on my own GPUs, fed by my optimizer and aware of the macro regime.

PythonLangGraphvLLMllama.cpp
Eight GigsEvolving

Local LLM Lab

A 9B model with 32k context entirely on an 8 GB laptop GPU, and a 27B model on a rented 48 GB GPU — behind the same OpenAI-compatible API.

llama.cppvLLMPythonDocker
Contract FirstIn production

Agent Harness & Context Engineering

How I work with coding agents on production data: read-only access, a memory vault, a knowledge graph and human-only steps.

Claude CodeMCPObsidianPython
Divergent SignalEvolving

Quant + AI Research Terminal

A web terminal where the optimizer says how much to hold, the agents say why — and the rows that matter are where they disagree.

PythonPostgresSupabaseReact
Living FrontierEvolving

Macro-Aware Portfolio Optimizer

A broken Markowitz spreadsheet rebuilt as a tested Python system that knows the macro regime.

PythonNumPySciPyscikit-learn
Two TracksIn production

Multi-Source Data Architecture

Every source behind the dashboards inventoried — what it is, where it lands, who owns it — and two tracks that never mix.

PostgresSQLSupabaseAWS Glue/Athena/S3
Self-ChainingIn production

Sync Engine: Legacy MySQL → Postgres Mirror

Edge functions that copy only what changed, one unit per call, with checksums, shadow tables and atomic publish.

PostgresSupabaseMySQLSQL
Triple RenameIn production

Legacy MySQL Forensics

Read-only investigations on a legacy MySQL whose failures were silent: the log said OK while the numbers went wrong.

MySQLSQLAWS
Gold LayerIn production

Canonical Marts on a Medallion Lake

My part of the company's data lake: canonical sales marts and the audit for moving dashboards off the legacy database.

AWS Glue/Athena/S3IcebergParquetSQL
InvariantsIn production

Data Reliability Suite

Invariants that cross the legacy DB, the mirror, the RPCs, the lake and the rendered UI — each one born from a real incident.

SQLPythonPlaywrightPostgres
Field GuideShipped

Data Atlas

A structural map of a company's data operation — systems, tables, pipelines, metrics and traps — written for whoever arrives next.

ObsidianSQLMySQLAWS Glue/Athena/S3
Exact AttributionClient work · In production

Performance-Marketing Reporting Backend

Daily ingestion from checkouts, ad tracker, click tracker and video analytics, exact cost attribution per video sales letter, and QA that blocks bad numbers.

PythonPostgresSQLSupabase
PostbackClient work · Shipped

Serverless Order & Campaign ETL

A Lambda pipeline from a checkout platform and an ad tracker into Postgres — idempotent, timezone-aware and recoverable.

PythonAWS LambdaPostgres
Ledger GlueClient work · Shipped

Finance Integrations

Corporate-card transactions into accounting, a multi-provider bank ETL with central-bank FX, and projection spreadsheets generated by code.

PythonAWS LambdaPostgresGoogle Sheets API
Fixed SeedShipped

End-to-End Synthetic Data Pipeline

A take-home challenge: synthetic data → pure-SQL DuckDB pipeline → Streamlit dashboard → the project's knowledge graph served over MCP.

DuckDBSQLPythonStreamlit
Born ClosedIn production

Auth, Roles & Invitations

Google sign-in limited to the company domain, roles per dashboard, invitations for people who never logged in, and a dev shortcut that can't open production.

SupabasePostgresReactTypeScript
Execute to PublicPartially remediated

Postgres RPC Exposure Audit

Most dashboard functions could be called by anyone holding the public key. Measured in production, revoked, and turned into an invariant.

PostgresSQLSupabase
Liquid GlassIn production

Analytics Dashboards Hub

Company reporting moved from a legacy BI tool into one React app of production dashboards.

ReactTanStackshadcnTailwind
Always ProjectedIn production

Executive Margin Cockpit

A live P&L down to contribution margin per order line, for leadership only — where net is always projected.

ReactTanStackPostgresSQL
Pipeline, Not BusinessIn production

Lead Processing Latency

How long a lead takes from checkout to SMS, e-mail and call center — exact p50/p95 per day, channel and event, and where leads get stuck.

ReactPostgresSQLMySQL
Effective RecoveryIn production

Lead Enrichment & ROI

Recovery and ROI of the queue that finds valid phone numbers for leads — where “phone found” isn't success.

ReactPostgresSQLTypeScript
Nothing RecomputableIn production

Tiered Affiliates Dashboard

One route, two readings by role — and no hidden number can be rebuilt from what's left on screen.

ReactTanStackPostgresSQL
Cell by CellIn production

AOV, Funnel & A/B

Average order value with and without upsells across gateways, funnel steps and A/B tests — migrated from an analyst's view with parity to the cent.

ReactRechartsPostgresSQL
Provider GrainIn production

Dispatch Match Rate & a Silent Outage

SMS and e-mail match rate measured at the provider's grain — and the audit that found a channel off while the dashboard stayed green.

SQLMySQLPostgresReact
Clone and GoIn production

Dashboard Scaffold & Local-First Workflow

A clone-and-go base for new dashboards, and a way to build them locally instead of paying per message in the app builder.

ReactTanStackshadcnTailwind
Inline ForecastClient work · Shipped

Financial Projection Studio

A web app for monthly financial projections next to embedded BI panels. I built the first version and later moved its data to embedded Metabase.

Next.jsReactFirebaseTypeScript
SajuEvolving

Four Pillars Oracle

A Korean Saju astrology web app — free chart, deep reading, optional account. Co-built; I owned the edge deployment.

Next.jsReactCloudflare WorkersSupabase
Eleven LayersShipped

Research Product Landing Page

The public page for an AI investment-research product: lead capture, sample reports and demo booking. I took it from first build to launch.

HTMLCSSJavaScriptCloudflare Pages