Kunal Kumar

Software Engineer

New Delhi, Delhi, India1 yr 9 mos experience

AI EnabledAI ML Practitioner

Key Highlights

Expert in fine-tuning state-of-the-art LLMs and VLMs.
Proficient in optimizing AI models for edge deployment.
Strong background in machine learning and computer vision.

Stackforce AI infers this person is a Machine Learning Engineer specializing in AI and Edge Computing.

Contact

Skills

Core Skills

Machine LearningAi SecurityEdge ComputingNatural Language Processing (nlp)Computer Vision

Other Skills

C++CLIPCalibration AlgorithmsData ScienceDeepStreamDynamic QuantizationGPU Programming(CUDA & Triton)Knowledge DistillationLLMsLORALow Level Optimization using CPPPyTorchPythonQLORAR&D for SOTA optimization algorithms

About

Hey there 🤗! I am Kunal, a Machine Learning Engineer. My areas of expertise/interests include: LLMs, VLMs, SentenceTransformers, CLIP, vLLM, llama.cpp, DeepStream, TensorRT, PyTorch, GPU Programming(CUDA & Triton), Low Level Optimization using CPP, Reinforcement Learning, and R&D for SOTA optimization algorithms Here's what I do as an MLE regularly: 1. Fine-tuning of LLMs & VLMs using SOTA fine-tuning techniques(lora, qlora, grpo, dpo) and defining custom evaluation metrics to benchmark model performances. 2. Serving fine-tuned model and optimizing it further for edge-based deployment using llama.cpp server engine in GGUF 3. Optimizing the object-detection tensorRT engine using calibration algorithms for edge devices like Nvidia Jestons 4. Developing multi-modal RAG for cross-camera reidentification of an object across multiple streams. 5. Documenting the experiment on white paper. I am a B.Tech graduate from Netaji Subhas University of Technology. I have a keen interest in Mathematics and Machine Learning, and I am open to collaborating on ML Research and Projects. Please feel free to contact me if my skills align with your objective: kunu5402@gmail.com

Experience

Javelin

Software Engineer

May 2025 – Present · 10 mos · San Francisco Bay Area · Remote

Working on Agentic AI Security Solutions, building guardrails for genAI application.

C++PythonNatural Language Processing (NLP)Computer VisionData ScienceMachine Learning+1

Bipolar factory

Machine Learning Engineer

May 2024 – Apr 2025 · 11 mos · Bengaluru, Karnataka, India · Remote

I work with edge devices like Nvidia Jetsons. Edge devices are best if data privacy is important because all the computations are done within the local network.
What do I do daily?
Finetuning various SOTA LLMs and VLMs for tasks like OCR, Action Recoginition, Object Indentification, NER, etc, using techniques like LORA, QLORA, etc.
Finetuning CLIP and VIT models for better image understanding for tasks like generating embeddings for cross-camera identification.
Optimizing the model performance further using techniques like dynamic quantization and knowledge distillation.
Optimizing tensorRT engine of object-detection models like yolo with techniques like calibration to accommodate more number of frames for inference.
Sometimes writing CUDA and Triton kernels for efficient matrix multiplication.
R&D for latest architectures and framework for efficient serving of LLMs and VLMs
The most important step is containerizing the end-to-end pipeline to avoid building the target with the same GPU instruction set repeatedly.