Second-year student at IIIT Bangalore. I like ML work with a clear question, a real experiment, and numbers you can check.
Things I care about right now:
- World models and model-based RL
- Multimodal learning (vision, language, control)
- Generative models and how we evaluate them
Happy to talk about research internships or AI/ML engineering roles.
Email me: lohithpasumarthi25@gmail.com
- Reward hacking in physics / structure-based rewards for 3D generation (ongoing)
- Reading and small repros around world models and multimodal agents
- Cleaner experiment and evaluation setups for ML projects
I only put finished or solid work in the project section below. Ongoing research stays light until it is ready.
|
PPO + ViT in PyBullet for continuous drone control under wind and clutter. 100% success, 0% collision. 8 ablations vs DDPG/SAC. Repo |
Privacy-preserving synthetic data (TabDDPM, TabSyn, TVAE, CTGAN + DP-SGD). 250K+ records, 94.6% downstream utility. IEEE DataPort Dashboard |
|
12-layer encoder (MLM) + generative decoder in PyTorch. SQuAD v2 58.2 F1. Live demo on HF Spaces. Demo |
YOLO + PaddleOCR for helmet / triple-riding checks and plate OCR. Test mAP50 about 0.92. Course team project. Repo |
More
- Job Trend Radar: Streamlit skill-gap / job-market tool
- Datathon 2025 winner at IIIT Bangalore
Languages
AI / ML
Tools
| lohithpasumarthi25@gmail.com | |
| lohith-pasumarthi | |
| GitHub | Lohith248 |
Best way to reach me is email.