Deepesh Kakkar

I build backend systems and AI pipelines that ship to production.

Global Winner at Qualcomm's Snapdragon Multiverse Hackathon 2026 .
What I've built and worked on:

Dragverse - phone-captured 3D scan → simulation-ready digital twin → robot control policy trained via PPO (Unity ML-Agents) → deployed to a physical robot through ONNX + INT8 quantization on a Hexagon NPU. Full capture-to-deployment pipeline, built in a two-day window.

AETHERIS - autonomous cyber deception platform on LangGraph + FastAPI that spins up adaptive sandboxes to lure and contain attackers, with WebSocket telemetry streaming and MITRE ATT&CK enrichment.
At Halliburton, shipped a 20+ page Next.js platform and cut deploy time from 18 minutes to 6 while sustaining 5–10 releases/week.

Edge AI inference pipeline for Qualcomm NPUs - 89.2% accuracy sustained at 2,671+ images/hour, hardware-tuned for real-time execution on resource-constrained devices. Invited by Qualcomm to build the production desktop version with their engineering team last year.

Final year @ TIET

GitHub: github.com/DEEPESH-845
Portfolio: deepesh.qzz.io