Selected work · 2024–2026

SYSTEMS BUILT
TO DO REAL WORK.

AI agents, security tooling, data systems, and hardware projects, documented around the problem, approach, and outcome.

Xternal AI · Product walkthrough

SEE HOW XTERNAL WORKS.

From one costly workflow to a production AI system your team can run, measure, and own.

Opportunity mapped

Find the work worth automating

Xternal starts with a fixed-scope opportunity map: understand the workflow, measure its baseline, and decide what AI should improve and what should stay human.

XTERNAL AI / 01Opportunity mapped
Timebox
2 weeks
Output
Prioritized roadmap
01AI Systems · Product EngineeringXternal

Xternal builds production AI systems around the workflows companies actually run. Engagements move from a fixed-scope opportunity map to a measured pilot, then scale validated systems with the integrations, controls, documentation, and training teams need to operate them confidently.

Challenge
Teams need AI systems that improve real workflows without adding another expensive platform or opaque experiment.
Outcome
Created an ownership-first path from one costly workflow to a production AI system the client can run and measure.

Decisions

  • Map the workflow and measure its baseline
  • Pilot the highest-value opportunity using real systems and data
  • Scale what works, then document and transfer ownership

Architecture

  1. Opportunity map
  2. Measured pilot
  3. Production integrations
  4. Controls and auditability
  5. Documented handover

What changed my thinking

  • Start with the workflow, not the model
  • A pilot needs a baseline to prove value
  • Documentation and handover are part of the product
AI AgentsAutomationFull StackIntegrationsAI Governance
02AI · Security · FounderAegis

Founded Aegis as an end-to-end AI-powered cybersecurity platform targeting the mid-market security gap. The platform combines autonomous OSINT pipelines, LLM-driven vulnerability analysis, and structured threat intelligence reporting — enabling security teams to dramatically reduce triage time and surface critical exposures before adversaries exploit them.

Role
Founder and product engineer
Challenge
Security teams need useful exposure context, not another queue of scanner output to triage by hand.
Outcome
Created an end-to-end platform direction that turns fragmented reconnaissance into reviewable threat intelligence.

Decisions

  • Combine reconnaissance and vulnerability context in one workflow
  • Use agents to structure findings instead of replacing analyst judgment
  • Design reports for both human review and downstream systems

Architecture

  1. Target scope
  2. OSINT and service discovery
  3. Vulnerability correlation
  4. Agent analysis
  5. Structured intelligence report

What changed my thinking

  • Explainability matters more than novelty in security workflows
  • Every automated conclusion needs traceable source evidence
aegisintelligence.org
PythonFastAPILLM AgentsOSINTPostgreSQL
View GitHub
03Data Engineering · InternshipFP&A Financial Dashboard

Engineered a unified FP&A application at Premier Early Childhood Education Partners that consolidates financial and operational data from disparate sources into a single source of truth. Automated recurring monthly reporting workflows and built interactive Power BI dashboards, eliminating over 15 hours of manual work per reporting cycle and improving decision-making speed for leadership.

Role
Data engineering intern
Challenge
Financial and operational reporting depended on disconnected sources and repeated manual reconciliation.
Outcome
Eliminated more than 15 hours of manual work from each monthly reporting cycle.

Decisions

  • Normalize inputs into a shared reporting model
  • Automate recurring transformations
  • Design dashboards around leadership decisions

Architecture

  1. Source systems
  2. Python and SQL transforms
  3. Snowflake model
  4. Power BI semantic layer
  5. Leadership reporting

What changed my thinking

  • A trusted definition is more valuable than a clever chart
  • Data quality checks belong inside the pipeline
PythonSQLPower BISnowflakeREST API
04AI · Automation · InternshipAI Program Management Agent

Developed an LLM-powered program management agent at Hendrickson that integrates with existing project tooling to automate task tracking, proactively surface blockers, and generate status summaries. The agent reduces manual coordination overhead for program managers and provides real-time project health insights through a conversational interface backed by FastAPI and PostgreSQL.

Role
AI engineering intern
Challenge
Program status lived across tools, making blockers easy to miss and summaries expensive to assemble.
Outcome
Built a working agent that surfaces blockers, tracks tasks, and generates project health summaries.

Decisions

  • Ground responses in existing project data
  • Separate retrieval from generated summaries
  • Keep the interface conversational while preserving source context

Architecture

  1. Project data
  2. Retrieval layer
  3. Agent workflow
  4. FastAPI service
  5. Program manager interface

What changed my thinking

  • Useful agents need narrow authority and visible provenance
  • Fallback behavior is part of the product
PythonLangChainFastAPIClaude APIPostgreSQL
05Full-Stack · NASA Space Apps 2025SpaceVitals Dashboard

Built for NASA Space Apps Challenge 2025 — a real-time astronaut health monitoring dashboard that ingests biometric data via Flask REST APIs and renders live visualizations using Chart.js. The system models astronaut vitals including heart rate, temperature, and oxygen levels, surfacing anomaly alerts to support mission health decisions.

Role
Full-stack engineer
Challenge
Turn multiple live biometric signals into a view that supports quick mission-health decisions.
Outcome
Delivered a working real-time health dashboard during NASA Space Apps 2025.

Decisions

  • Model vitals behind a small REST interface
  • Separate ingestion from visualization
  • Make anomalies visually distinct from normal telemetry

Architecture

  1. Biometric inputs
  2. Flask API
  3. Vital normalization
  4. Alert rules
  5. Live dashboard

What changed my thinking

  • Operational dashboards must prioritize exceptions
  • Stable data contracts make rapid frontend iteration possible
FlaskJavaScriptREST APINASA APIChart.js
View GitHub
06Security · AutomationAutonomous Recon & Vuln Scanner

The core intelligence engine of Aegis — an autonomous pipeline that chains OSINT enumeration, Nmap-based port and service discovery, and CVE database correlation to produce structured threat intelligence reports. Designed to run headless on target domains and deliver actionable findings in both machine-readable JSON and human-readable markdown formats.

Role
Security engineer
Challenge
Reconnaissance data arrives from separate tools and lacks a consistent evidence trail.
Outcome
Built the core intelligence pipeline used by Aegis for structured reconnaissance.

Decisions

  • Chain enumeration, service discovery, and CVE correlation
  • Preserve raw evidence beside normalized findings
  • Export for humans and machines

Architecture

  1. Authorized target
  2. OSINT enumeration
  3. Nmap discovery
  4. CVE correlation
  5. JSON and Markdown reports

What changed my thinking

  • Scope validation must happen before scanning
  • Confidence and evidence should travel with every finding
PythonNmapFastAPIOSINTSQLite
07Hardware · IoTMulti-Sensor Health Tracker

Designed and built a wearable multi-sensor health tracking system combining a Seeed Xiao microcontroller and Raspberry Pi. Wrote C/C++ firmware for real-time acquisition of GPS coordinates, heart rate, and ambient temperature over UART, then streamed data to a Python-based dashboard for live visualization and historical analysis.

Role
Embedded and application engineer
Challenge
Acquire several sensor signals reliably on constrained hardware and make them useful off-device.
Outcome
Produced a working hardware-to-dashboard prototype for live health telemetry.

Decisions

  • Keep firmware responsibilities small
  • Stream readings over a simple serial contract
  • Visualize current and historical measurements separately

Architecture

  1. GPS and health sensors
  2. Seeed Xiao firmware
  3. UART transport
  4. Raspberry Pi service
  5. Python dashboard

What changed my thinking

  • Simple protocols are easier to debug at the hardware boundary
  • Sensor failures need explicit states
C/C++Raspberry PiIoTPythonUART
08Web AppEmployee Scheduling System

Built a full-stack employee scheduling application with role-based access control, shift management, and integrated team communication. The system replaced manual scheduling spreadsheets, reduced scheduling conflicts by introducing real-time conflict detection, and improved overall team coordination efficiency by 30%.

Role
Full-stack engineer
Challenge
Manual scheduling made conflicts and team communication difficult to manage.
Outcome
Improved team coordination by 30% while replacing spreadsheet-based scheduling.

Decisions

  • Model roles and shifts explicitly
  • Detect conflicts before saving
  • Keep schedule and communication in one workflow

Architecture

  1. React interface
  2. Node.js API
  3. Role checks
  4. Firebase data
  5. Schedule notifications

What changed my thinking

  • Authorization belongs at the data boundary
  • Conflict states should explain how to recover
ReactNode.jsFirebaseREST API
09AI · Product EngineeringBloomGuard

Contributed across design, engineering, and user experience on BloomGuard. The project reflects a cross-functional approach to building AI-enabled products, connecting technical implementation with a clear and usable product experience.

10Data · Financial IntelligenceFinSight

Built FinSight around agent-driven financial intelligence workflows, contributing across data pipelines, engineering, and product experience. The work focused on turning complex financial information into a more structured and usable system.

11Security · SystemsSecure Enterprise Storage

Worked on a secure enterprise storage system with an emphasis on protecting data and maintaining reliable system behavior. The project joined security thinking with practical systems engineering for an enterprise environment.

12AwardCMU NexHacks '26 · Seda AI Researchathon

Placed 1st at Carnegie Mellon's NexHacks '26 hackathon and won the Seda AI Researchathon — both recognized for building high-impact, production-quality AI solutions under tight time constraints. The NexHacks project applied autonomous AI agents to a real-world problem domain; the Seda win focused on novel research into AI-driven cybersecurity threat modeling.