I’m Jeswanth Mukesh, an AI engineer based in Chennai, India.

I’m a Member of Technical Staff at ZLabs, Zohocorp — I train LLMs, experiment with architectures, build custom evaluations, read papers, and try to apply what I learn.

Research interests

Multimodal Learning — integrating text, vision, and audio into unified representations for tasks like document intelligence and multimodal retrieval.

Inference Speed Optimization — low-level GPU kernel engineering (GPU Mega Kernels, cross-platform kernels with wgpu), speculative decoding, and constraint decoding to reduce latency across hardware backends.

Reinforcement Learning & Interpretability — RL for training and aligning large language and multimodal models, and mechanistic interpretability to understand model behavior.

Experience

ZLabs, Zohocorp — Member of Technical Staff · May 2024 – Present

Pinaca Technologies — Junior Machine Learning Engineer · Sep 2023 – May 2024

  • ML solutions for text classification (AML project) using BERT
  • Redis database migration, reducing latency by over 200×
  • Skills: PyTorch, Redis, LlamaIndex, Qdrant, PyTorch Multiprocessing

Pinaca Technologies — Junior Machine Learning Intern · Jul 2023 – Sep 2023

  • Face recognition with improved resolution using OpenCV Super Resolution

FaceTagr — Software Development Intern · May 2021 – Jun 2023

  • CRM and ML application interfaces; microservice architecture with Nginx load balancing
  • Stack: Django, MongoDB, Docker, AWS, Keycloak, Nginx, MinIO

Projects

Contrastive Audio-Text Embedding for Fast Multi-Modal Retrieval Text embedding model trained with Whisper-large and CLIP-style contrastive loss on GigaSpeech. Achieved NDCG@5 of 81% in validation.

Object Detection and Navigation Assistance Using LiDAR and Stereoscopic Camera Research on using Intel RealSense D435i + L515 with varying focal lengths to assist blind individuals in navigation.

CopyTxT Online React + FastAPI app for storing and retrieving code snippets via a 4-digit key. Syntax highlighting for 20+ languages, GitHub Actions CI/CD, MongoDB, deployed on AWS EC2.

Ecommerce-App using Vue.js and Flask Full-stack e-commerce app with role-based auth, Redis caching, and Celery for scheduled reports.

Education

Vel Tech High Tech Dr.Rangarajan Dr.Sakunthala Engineering College B.Tech in Artificial Intelligence and Data Science · CGPA: 8.17/10.0 · 2020 – 2024

Skills

Languages: C/C++, Python, Java, JavaScript, CUDA (Beginner)

Technologies: PyTorch, Transformers, Torch Distributed, Slurm, Hugging Face, vLLM, SGLang, Git, Aspose.pdf

Find me