name: Stephen Sookra
role: Software Engineer Intern @ XR Dojo
education: Kennesaw State University · CS · AI/ML Concentration · 3.8 GPA · '28
record: 12 hackathon awards across 10 events
latest: 1st place · Technologists of Color Chaos Challenge (Parliament, Aug 2026)
building: Manifest · Curtail · Attest · ATL Quest
learning: Trustworthy AI · Multi-Agent Governance · LLM Inference on Arm · On-Device Models
mentors: Aaron Butler · Henry
mission: Founding a company in trustworthy AI for high-stakes domains
location: Atlanta, GA (UTC-5)
contact: stephensookra@gmail.com · stephensookra.com- Latest win 1st place at the Technologists of Color Chaos Challenge with Parliament, a one-hour build for Grant Park, alongside Khadim Drame, Ryann Ridgeway, and Darren L. Martin
- Recent wins Best Use of Gemma 4 for Throughline at Hack RenderATL · Most Innovative Project ($750) for AccessGate at the IBM AI Builders Challenge · $5,000 Overall Grand Prize + Most Innovative for VARSITY across the IBM SkillsBuild Sports Challenges
- In judging Erasure-Proof (CockroachDB x AWS) · GravitonKV (Arm Create 2026) · Aloud (YouCam Skin AI) · Silent-Drift Sentinel (DataHub)
- Now building Manifest, a regulatory critical-path planner for university CubeSats · Curtail, a governed multi-agent fleet for California water law · Attest, a phone agent that refuses to guess · ATL Quest (AR + XR @ XR Dojo)
- Summer 2026 taught web + game dev as a SMASH Academy TA (Morehouse / Spelman)
- Ask me about hackathons · AR/XR · AI/ML · founding · Atlanta tech scene
Hackathon record · 12 awards across 10 events · 🏆 $5,000 Grand Prize @ IBM SkillsBuild · 💡 Most Innovative @ IBM SkillsBuild June · 🏆 Best Use of Gemma 4 @ Hack RenderATL · 💡 Most Innovative @ IBM AI Builders July · 🥇 1st @ ToC Chaos Challenge · 🥇 1st of 22 @ Actian Build Challenge · 🥇 1st @ HMI 2026 · 🥈 2nd of 4,700+ @ Wells Fargo / GCA · 🥈 2nd + 🥉 3rd @ KSU Social Good · 🥉 3rd @ KSU FinTech · 🥉 3rd @ Vibra ATL
|
🏆 VARSITY Grand Prize + Most Innovative · IBM SkillsBuild Real-time, screen-reader-native AI that explains VAR and offside calls to blind football fans, grounded in the IFAB laws. Python · IBM Granite · RAG · a11y |
🏆 Throughline Best Use of Gemma 4 · Hack RenderATL Record-integrity layer: reconciles what one institution asserts against independent authorities, 51.9% divergence across 1,344 entities, and suppresses its own largest rule when a source fails the coverage gate. Python · Gemma 4 · FastAPI · React |
|
💡 AccessGate Most Innovative · IBM AI Builders July Caption + audio-description conformance pre-check for film: 23 cited rules with a gated generative fix on IBM Granite. Every cited clause is a verbatim substring of the committed corpus. Python · IBM watsonx · React · a11y |
🥇 StepSafe 1st · HMI 2026 Computer-vision diabetic foot ulcer triage from a single photo: expert-level assessment for any clinician. TensorFlow · MobileNetV2 · FastAPI · React |
|
🥈🥉 PyroLens 2nd + 3rd · KSU Social Good AI prescribed-burn decision support fusing IoT sensors, weather, and satellite data for safer controlled burns. Python · scikit-learn · ESP32 · Supabase |
🥇 Parliament 1st · ToC Chaos Challenge One-hour chaos build: a one-scroll field school for Grant Park, Atlanta's oldest park at 131.5 acres, with a field pass that actually enrolls you. HTML · CSS · JS · Supabase |
|
🔐 Erasure-Proof CockroachDB x AWS Provable crypto-erasure for agent memory: one SERIALIZABLE transaction shreds the embedding, keeps a hash-chained log, and anchors a KMS-signed proof to S3 Object Lock. CockroachDB · AWS KMS · S3 Object Lock · Python |
💧 Curtail Google x Devpost Agentic Governed multi-agent fleet for California water law: watches the river, computes who must stop diverting under priority, drafts the order a human official signs. Python · multi-agent · water law · Postgres |
|
📋 Renova Build with Gemma Offline, on-device Medicaid renewal assistant: reads a renewal notice and returns the deadline, case number, and documents in plain English. Five states. Gemma · ONNX Runtime Web · PWA · iOS |
⚡ GravitonKV Arm Create 2026 KV-cache quantization tradeoff study for LLM inference on AWS Graviton4, with a reproducible harness and a canonical results schema. llama.cpp · KleidiAI · Graviton4 · arm64 CI |
More on DevPost · stephensookra.com
| AI / ML |
|
| Languages |
|
| Web / Mobile |
|
| Data / Cloud |
|
| Tools |
|
Credit: Inspired by @tylinndd · originally adapted from @1010nishant
Built by Stephen Sookra · Atlanta, GA · shipping ambitious things
Last edited: August 2026
Built by Stephen Sookra · Atlanta, GA · shipping ambitious things
Last edited: August 2026