Project Index
Projects across AI Agent/MCP, systems engineering, intelligence products, web analytics, data tooling and long-running Python engineering.
The engineering journey itself is the artifact: architecture refactoring, benchmarks, regressions, observability and AI-human collaboration drive a long-running QUIC/network data-plane effort toward convergence.
Not a QUIC performance leaderboard, but a production-engineering case showing how AI-assisted hypotheses are measured, falsified, refactored and converged.
Active, stable and paused describe the current lifecycle of each project. Paused projects remain working artifacts while expansion is deprioritized in favor of deeper iteration.
A quick-start personal intelligence engine that collects, organizes and filters information across jobs, schools, industries, markets and personal projects, turning scattered searches into actionable lists.
A Rust gateway built to replace nginx, with HTTP/1.1, HTTP/2, WebSocket and gRPC lanes, bounded relay lifecycles, integrated TLS certificate management, security rules and observability.
An industry-intelligence radar for consumer brands, channel monitoring and market-entry decisions, continuously turning news into brand moves, channel signals, market opportunities and operating implications.
A multi-topic intelligence dashboard for lightweight research, organizing weather, food safety, Greater Bay Area news, automotive markets, agriculture, tenders and site-selection signals into fast-scanning briefings.
A job-intelligence tool for opportunity discovery, data cleaning, role analysis and resume-positioning support, gradually automating collection, analysis, presentation and publishing.
An AI work-asset system for individuals and teams that organizes prompts, skills, project context, tool instructions, failure conditions and version history into structured resources that Agents can search, invoke, reuse, version and access under authorization.
A deterministic, non-LLM Python article extractor that reports complete-body extraction for about 98% of its target site set and serves as an ingestion layer for Agent/LLM web workflows.
Structured logging for Python services and data pipelines, with request-scoped context, text/JSON output, rotating files and optional bounded queue delivery.
A Python JSON toolkit with an orjson-first backend, serialization for complex types such as datetime, Decimal and UUID, circular reference handling and an explicit standard-library fallback.
A QUIC proxy and transport engineering project focused on connection scheduling, congestion behavior, recovery, forwarding stability and controllable operation under stress; current work concentrates on protocol boundaries, scheduling, benchmarking and observability.
A multi-tenant analytics product for lightweight measurement, campaign templates and analysis, in the same problem space as web analytics; current work focuses on data structures, metric definitions, tenant boundaries, content-source integration and explainable review.
A network runtime for domain rules, gateway routing, traffic decisions and recovery scenarios, with current work focused on rule observability, decision traces, fault recovery, configuration convergence and cross-platform operation.
A site-selection assistant for store owners, expansion teams and regional operators, providing location checks, trade-area profiling, competitor density, accessibility and time-of-day observation guidance before on-site visits.
Agent runtime, MCP/tool integration, task state and execution boundaries.
An MCP engineering work ledger and coach that preserves tasks, decisions, risks and validation evidence across Agent sessions and context resets, with project memory, bounded recovery and deterministic governance.
An MCP server for UML/DFD and architecture modeling, with revision history, deterministic validation, impact analysis and refactor planning. Context packs restore goals, decisions, open issues and the focus subgraph across Agent sessions.
Rust Responses-to-Chat protocol adapter for Codex CLI.
Secure embedded AI conversation infrastructure; also used by this portfolio.
A deterministic, non-LLM web-content extraction layer that provides clean inputs for Agent/LLM web research, retrieval and context construction.
Enterprise document extraction, production ETL and governed MCP database access: from data pipelines to explicit Agent tool boundaries.
Archive inspection, member reads and extraction for document ingestion and ETL, supporting nested archives, directory batches, password policies and JSON reports across ZIP, 7z, RAR and TAR.
Structures DOCX/PDF contract tables for audit and evidence workflows, including header detection, similar-header merging, field mapping and complex multi-table documents; historical materials report about 95% header-identification accuracy.
A Python data-cleaning and governance track covering orchestration, chunked processing, primary-key deduplication, dirty-data side paths, quality validation, load auditing and failure recovery, extending broader Airflow/DataX production experience.
MySQL and PostgreSQL access through governed MCP endpoints: Connection → scoped Endpoint → token-based ACL. Table allowlists, read-only SQL AST checks, audit logs and usage metering keep database tools within explicit boundaries.
Securities metadata, indexing and retrieval infrastructure.
A reproducible synthetic OHLCV generator for demos, smoke tests, documentation examples and fast validation of market-data pipelines.
A Polars-based A-share backtesting framework covering T+1 execution, market constraints, order state transitions, risk controls, dynamic slippage and performance analysis.