Thales Savkli Résumé

Applied data science · Software engineering

Thales
Savkli

I build autonomous and data-intensive systems: 4D path planners, real-time telemetry pipelines, and simulation platforms for critical infrastructure.

Virginia Tech CMDA graduate, two summers at Johns Hopkins APL, and a submitted manuscript in coding theory.

  • B.S. CMDA · VT 2025
  • IN-MAJOR GPA 3.97
  • JOHNS HOPKINS APL 2024 & 2025
Download résumé Get in touch
Portrait of Thales Savkli
SAVKLI, THALES · FIG. 01
FIG. 02 · DUBINS PLANNER

A fixed-wing glyph flying shortest turn-straight-turn routes around drifting no-fly zones, replanning whenever the route is cut off. Same turn-constrained geometry as my 2024 path-planning work, in a few hundred lines of vanilla JavaScript with no libraries. Click or tap the panel to replan to a new target.

WP-01 · ROUTE LOG

Experience

  1. AUG 2022 · MAY 2025  ·  BLACKSBURG VA

    Virginia Tech · B.S. Computational Modeling & Data Analytics

    Data-science degree with minors in Computer Science, Mathematics, and Statistics. In-major GPA 3.97 / 4.0.

  2. MAY 2024 · SEP 2024  ·  LAUREL MD

    Johns Hopkins APL · Software & Algorithms Intern

    Fixed-wing UAS path planning and simulation

    Built a 4D path-planning environment for fixed-wing aircraft with dynamic obstacles and Dubins paths. Designed the performance tests and algorithm validation under time-based airspace restrictions and multi-drone coordination, working with subject-matter experts to refine the APIs and UX. Also built a Python desktop app that automated task creation and orchestrated workflows, achieving a 50%+ speedup.

  3. MAY 2025 · SEP 2025  ·  LAUREL MD

    Johns Hopkins APL · Software & Algorithms Intern

    Infrastructure modeling & simulation · Unmanned traffic management

    Returned to build CityLink, a Python simulation platform for critical-infrastructure scenarios that supports faster incident-response analysis. Wrote a rule-based unmanned traffic management engine that flags airspace and procedural violations from live drone telemetry, on a Dockerized Kafka pipeline that ingests, evaluates, and responds in real time. Built AI-assisted engineering workflows with Claude Code, on-prem LLM frameworks, and MCP tooling.

  4. ONGOING

    Continued consultation with both APL teams after each internship term.

WP-02 · SELECTED WORK

Selected work

Four systems in depth. The diagrams are abstract sketches of the underlying ideas, not the real interfaces.

JOHNS HOPKINS APL · 2024 · FLAGSHIP

4D Fixed-Wing Path Planning

A planning environment that routes fixed-wing aircraft through space and time, around obstacles that move. A fixed-wing aircraft cannot stop or pivot in place, so every maneuver is built from straight tangents and minimum-radius arcs: Dubins paths. I designed the performance tests and algorithm validation that stressed the planner under time-based airspace restrictions and multi-drone coordination, and worked directly with subject-matter experts to refine the planning APIs and interface so the tool held up in practice, not just on paper.

  • Python
  • Dubins-path planning
  • Simulation

INTERNAL TOOLING · JOHNS HOPKINS APL

R TANGENT START POSE GOAL POSE SHORTEST OF SIX CANDIDATES TURN · STRAIGHT · TURN
FIG. 03 · ARC-TANGENT-ARC CONSTRUCTION

JOHNS HOPKINS APL · 2025

Real-Time UTM Engine

A rule-based unmanned traffic management engine that watches live drone telemetry and flags airspace and procedural violations as they happen. It runs on a Dockerized Kafka streaming pipeline that ingests telemetry, evaluates it against the rule set, and triggers a response in real time.

  • Python
  • Kafka
  • Docker
  • Streaming telemetry

INTERNAL TOOLING · JOHNS HOPKINS APL

INGEST RULE 01 RULE 02 RULE 03 EVALUATE RESPOND FLAG TELEMETRY IN · RULES APPLIED · VIOLATIONS FLAGGED EACH STAGE A CONTAINER · STREAMS IN BETWEEN
FIG. 04 · INGEST · EVALUATE · RESPOND

VIRGINIA TECH · NSWC DAHLGREN · 2025

Documetrics

Documentation-quality metrics for source code: conciseness, completeness, accuracy, and comment density, computed with UniXcoder and Sentence-BERT embeddings. The project compared human-written and LLM-written documentation to surface recurring patterns and concrete recommendations, with an interactive dashboard as the entry point to the analysis.

  • Python
  • UniXcoder
  • Sentence-BERT
  • React
  • Tailwind CSS
  • Chart.js
View repository
HUMAN-WRITTEN LLM-WRITTEN COSINE SIM EMBEDDING SPACE · 2D SKETCH
FIG. 05 · EMBEDDING-SPACE SKETCH

WP-03 · RESEARCH

Research & manuscript

LEAD RESEARCHER · CODING THEORY & CRYPTOGRAPHY · VIRGINIA TECH · 2023 TO 2025

Rank-metric codes and Binomial Moment Determined codes

I constructed families of rank-metric codes with targeted algebraic properties and formalized the theoretical conditions that govern them. MAGMA experiments validated the results, benchmarked candidate constructions, and analyzed structural invariants. I led the workflow from mathematical formulation through computational validation, writing, and revision with faculty guidance.

The result is a co-authored manuscript on Binomial Moment Determined (BMD) codes, covering their construction, existence, and cryptographic applications. The manuscript is submitted; no public preprint yet.

MAGMA · RANK METRIC · CONSTRUCTION AND EXISTENCE

WP-04 · SKILLS

Technical skills

Skills with a blue marker are used in work on this page. Hover or tap one to see where.

LANGUAGES

  • Java
  • C
  • R

DATA & SCIENTIFIC COMPUTING

  • Statistics
  • CUDA

SYSTEMS & CLOUD

  • Git
  • AWS
  • DynamoDB

 

I also build LLM-assisted engineering workflows: Claude Code, on-prem LLM frameworks, and MCP tooling.

WP-05 · ROUTE COMPLETE

Get in touch

Open to software engineering, applied data science, and machine learning roles. Email is the fastest way to reach me.