Pranoy Radhakrishnan

Machine Learning Engineer

Trivandrum, Kerala · pranoyalkr@gmail.com · GitHub · LinkedIn

Open Source Contributions

  • ViT-PyTorch — Merged PR implementing JetViT and ViT-5.
  • Transfusion-PyTorch — Implemented unconditional KV caching, matching null-token sequence length to prompts for correct CFG alignment, and adding classifier-free guidance.
  • Rectified Flow-PyTorch — Implemented the Unconstrained Alignment (UA) Flow and LapFlow.
  • Mimic-Video — Improved noise scheduling, VAE input normalization, and action decoding with proprioceptive masking.
  • MetaController — Fixed Switching Unit logic by incorporating the previous subgoal (zt-1) for correct hierarchical control.
  • Pi-Zero-PyTorch — Fixed CFG RoPE alignment, temporal buffering, device compatibility, and unconditional pass robustness.

Personal Projects

  • CNN–LSTM (Video Classification) — CNN feature extraction combined with LSTM temporal modeling for video classification.
  • End-to-End Multi-Instance Pose Estimation with Transformers — DETR-style end-to-end trainable architecture for multi-person pose estimation without post-processing.
  • Large-scale Visual Relationship Understanding — Scene graph generation for detecting and classifying visual relationships between objects at scale.
  • Attention Models — Clean PyTorch implementations of SOTA attention mechanisms and transformer variants.
  • Scene Graph ViT — Scene graph generation using Vision Transformers for structured visual understanding.
  • Pix2Seq — Object detection framed as a sequence generation task following the Pix2Seq framework.
  • D2iT — Diffusion model built on a Transformer backbone (DiT-style) for image generation.
  • LSTM Object Detection — LSTM-SSD for temporal object detection across video frames.

Hover over a project
to preview it here

Career Timeline

Mar 2026Present
Avataar AI Bangalore
AI Engineer
Feb 2025Feb 2026
Aqxle AI Bangalore
Senior AI Research Engineer
Apr 2024Jan 2025
Accessia Bangalore
Computer Vision Engineer
Nov 2021Feb 2024
First Sense Bangalore acqui-hired → Accessia
Computer Vision Engineer
Aug 2018Jan 2021
Neuroplex Trivandrum
Co-Founder & Computer Vision Engineer
Aug 2017Dec 2017
CareStack Trivandrum
Software Engineering Intern