Kia Pousti Get in touch

Software engineer · San Diego, CA

Kia Pousti

Fast graph systems. Practical AI.

I build routing engines in C++ and applied AI in Python. Run your cursor through my name, then scroll: the same dots rebuild into what I work on.Scroll, and the dots in my name rebuild into what I work on.

Scroll

About

I think in graphs.

A math and computer science degree, applied to code that has to run fast on real hardware.

I studied Mathematics and Computer Science at UC San Diego, with a minor in Sociology, and graduated in June 2026. Today I am a software developer at Solar Turbines, a Caterpillar company, in its IT rotational program.

The work I enjoy most is graph algorithms, routing and data layouts. I wrote an A* engine in C++ that routes over about 1.2 million road nodes.

I also build applied AI: anomaly detection prototypes with tests and case studies, and LLM agents that call tools to answer real questions. Every project on this page runs a small live demo.

  • 1.2Mroad nodes in my A* routing engine, with about 2.5M edges
  • 89%smaller route payloads after polyline simplification
  • 0.22 → 0.47fault recall, baseline versus Isolation Forest
  • PR #2973Tarjan's algorithm contributed to TheAlgorithms/C++

Experience

Manual work, turned into software.

I build the apps and automations that give engineering and business teams their hours back.

Aug 2026 to now

Software Developer, IT Rotational Program

Solar Turbines, a Caterpillar company · San Diego, CA

  • Selected for a competitive rotational program, building and supporting enterprise apps for engineering and business teams.
  • Automated manual business processes with Power Apps and workflow automation, saving the team hours of work each week.
  • Built and shipped internal websites, using GitHub Copilot across several AI models to move faster.
  • Tracked and delivered work through ServiceNow and Azure DevOps, analyzed data in Power BI, and wrote PowerShell scripts in a sandbox environment.
  • Power Apps
  • GitHub Copilot
  • PowerShell
  • Power BI
  • ServiceNow
  • Azure DevOps
  • Visio
Automation one task, two waysillustration
  • Power AppsHours saved every weekManual business processes replaced with apps and automated workflows.
  • GitHub CopilotAI-assisted buildsInternal websites shipped faster by working across several AI models.
  • PowerShellSandbox methodScripts are written and run in a sandbox environment.
  • Visio · ServiceNow · Azure DevOpsDocumented and trackedProcess and system diagrams for stakeholders, with work delivered through tickets and boards.

June 2026

B.S. Mathematics–Computer Science

University of California, San Diego · Minor in Sociology

Coursework in data structures, algorithms, theory of computation, discrete math, probability, numerical analysis and computer organization.

  • Algorithms
  • Probability
  • Numerical analysis
  • Theory of computation

Projects

Four projects, all running live.

One project per screen. Scroll, and the dots behind rebuild into each one.

Inside Route Engine

One request, start to finish.

  1. 01 SnapA raw latitude and longitude is matched to the nearest road node.
  2. 02 SearchA* explores outward, steering by the straight-line distance still to go.
  3. 03 RouteThe cheapest path is rebuilt by walking back from the goal.
  4. 04 SimplifyPoints the map does not need are dropped. In the real engine, that cuts route payloads by about 89%.
A* shortest pathmove your cursor over the map

Route Engine

View code ↗

An A* routing engine in modern C++ over real OpenStreetMap data, about 1.2M nodes and 2.5M edges, with a haversine heuristic for geographic cost.

  • Cache-efficient Compressed Sparse Row graph layout to cut memory use and speed up traversal.
  • Spatial snapping from raw latitude and longitude to the nearest road node.
  • REST API serving GeoJSON routes to a Leaflet map, with payloads about 89% smaller after polyline simplification.
  • C++
  • A*
  • CSR graphs
  • REST
  • GeoJSON
Telemetry anomaly detectionillustration, synthetic data

Industrial AI Notebook

View code ↗

A public library of applied-AI prototypes for industrial IT problems. Each one ships with a case study, a pytest suite and GitHub Actions CI, and uses synthetic data only.

  • Benchmarked a rolling z-score baseline against an Isolation Forest on sensor telemetry.
  • The model doubled fault recall, from 0.22 to 0.47, and caught slow-drift failures the baseline missed.
  • Python
  • scikit-learn
  • pandas
  • pytest
  • GitHub Actions
Dijkstra settling nodes by distancenumbers are path costs

Mini Maps, a weighted shortest-path engine

View code ↗

An adjacency-list graph engine with O(V + E) traversal across weighted and unweighted networks.

  • BFS and binary-heap Dijkstra, O(E log V), with deterministic path reconstruction.
  • Modular components and a reproducible Make build.
  • C++
  • Dijkstra
  • BFS
  • Make
Tarjan strongly connected componentseach color is one component

Open source: TheAlgorithms/C++

View PR ↗

Contributions to a widely used open-source algorithms library, refined through peer code review.

  • Tarjan's strongly connected components algorithm, PR #2973.
  • Stoer–Wagner global minimum cut, O(VE + V² log V), with assertion-based tests.
  • C++
  • Tarjan SCC
  • Stoer–Wagner

Side builds

Built for my own daily use.

Seven personal projects, some finished and some still in progress. The screens show sample data.

Agent City

Agent system

A set of personal agents, each with one job, pictured as a lagoon of overwater huts. They ride in to one hub, and the dashboard surfaces only what needs a decision from me.

  • Agents
  • Scheduling
  • Dashboard

Surf Alert

Daily alert

A morning notification on whether the surf is worth the paddle, built on public wave, wind and tide data.

  • Automation
  • Open-Meteo
  • NOAA tides

How is the surf tomorrow morning?

→ surf_forecast · tides

Three to four feet with light offshore wind. Best window is 6:30 to 8.

Kia-bot

Voice agent

A hands-free voice agent for the car. A phone shortcut sends my question to a FastAPI backend, where Claude picks a tool and answers out loud.

  • Python
  • FastAPI
  • Claude API
  • Tool calling

Quant Backtest Lab

Research

A paper-trading framework for a trend strategy, with walk-forward validation and trading costs modeled. Results are reported as they are.

  • Python
  • Walk-forward
  • Cost modeling

Lead Autoresponder

Automation

Reads each inquiry to a small business, classifies intent and urgency, and replies right away. Complaints always go to a person.

  • Python
  • LLM classification
  • Human in the loop

Market Debrief

Dashboard

A daily page for my own portfolio: live prices, upcoming events for each holding, and a signal when a pullback from the 52-week high crosses a set tier.

  • JavaScript
  • Python
  • Twelve Data API

Math 163 Study Game

Study tool

An interactive study guide I made for a history of mathematics course: flashcards, matching games and quizzes with XP, from Greek geometry to Renaissance algebra.

  • HTML
  • JavaScript
  • Quiz engine

Skills

What I build with.

What I reach for, grouped by the job it does.

  • LanguagesC++ (STL, memory management), Python, JavaScript
  • AlgorithmsA*, Dijkstra, BFS, Tarjan SCC, Stoer–Wagner min-cut, complexity analysis
  • Data structuresCompressed Sparse Row, adjacency lists, heaps, trees, hash maps
  • ML and datascikit-learn, pandas, NumPy, Isolation Forest, precision and recall
  • SystemsMemory locality, cache-aware data layout, input validation
  • BackendREST APIs, HTTP, GeoJSON
  • EnterprisePower Apps, Power BI, ServiceNow, Azure DevOps, PowerShell, Visio, GitHub Copilot
  • ToolsGit, GitHub Actions, pytest, CMake, Make, Linux

Contact

Want to build something, or talk about a role?

I am based in San Diego. Email is the fastest way to reach me.

Off the clock

Away from the keyboard.

  • Mechanically inclinedHelped design and integrate an ESC, the speed controller, for an electric skateboard to raise its top speed and efficiency. I also sourced and mounted a replacement motor in my car myself, a job that runs thousands at a shop.
  • SportsBasketball, soccer and volleyball.
  • BoardsSurfing and skating.
  • CarsDetailing by hand: clay bar, polish and ceramic coating.
  • FamilyI am a family guy.
  • ReadingSelf-improvement books, such as Future Proofing You by Jay Samit and Man's Search for Meaning by Viktor E. Frankl.
  • MarketsI invest in the market, with positive returns so far, and I am working toward a career in the quant field.