🔭 AI Architect @ BLUE — building a real-time, multi-camera VLM pipeline for warehouse & retail vision analytics.
🧠 I work across LLMs, VLMs, multi-agent systems, and on-device inference — taking models from research to production at the edge.
🤗 Publishing open-source MLX-quantized LLMs on HuggingFace, making frontier models runnable locally on Apple Silicon.
Previously:
💻 Founding AI Engineer @ Stealth — a 27-agent orchestration framework + multi-stage document enrichment pipeline for AI-powered factory floor planning.
💻 Senior AI Engineer @ Avathon — scaled a CV platform to 600+ cameras across enterprise deployments, and added LLM + VLM layers to a pure-CV stack.
💻 Deep Learning Engineer @ TCS — shipped CV models for Smart Mobility & automotive data pipelines.
🛰️ Started out interning at ISRO's Regional Remote Sensing Centre.
🎓 B.Tech in Computer Science & Engineering, 2021.
💬 Reach out for projects, collabs, or just an interesting discussion.
- ml-explore/mlx-lm - Add sanitize method to Granite model for tied embeddings
- Ultralytics/Yolov3 - Fix ONNX inference code
- Ultralytics/Yolov5 - Fix FP32 TensorRT model export
23 MLX-quantized & uncensored LLMs — making frontier models runnable locally on Apple Silicon.
- MiniCPM5-1B-Uncensored
- LFM2.5-8B-A1B-Uncensored
- mellum2-12b-a2_5b-thinking-optiq-5bpw-mlx
- locateanything-3b-mxfp4-mlx
- granite-4.1-8b-mxfp4-mlx (MLX quants of the model I contributed Granite tied-embedding support for in mlx-lm)
- … see all 23 models →
A few favorites — full archive on Medium.
LLMs & Fine-Tuning
- Merged > Fine-Tuned? A Case Study on Qwen3 and Domain Fusion
- Fine-Tuning LLMs for Refusal
- Common Questions While Working with Large Language Models
Edge AI & Deployment
- Run YoloV5s with TensorRT and DeepStream on Nvidia Jetson Nano
- GPU, CUDA & Accelerated Programming using Numba in Python
- Getting Started with GStreamer in Python
Paper Summaries
- torch.manual_seed(3407) is all you need
- MetaFormer is Actually What You Need for Vision
- RepVGG: Making VGG-style ConvNets Great Again
→ Read all my articles on Medium
