AKASHJAMES

Vision, voice, language — at scale, on the edge, and everywhere in between.

VERONICA · AI COPILOT · STANDBY
V
AGENTSMULTIMODALVISIONEDGECUDAAGENTSLLMSDEEPSTREAMRAGJETSONTENSORRTAGENTSMULTIMODALVISIONEDGECUDAAGENTSLLMSDEEPSTREAMRAGJETSONTENSORRT
MANIFESTO · 001
I design AI that behaves less like a feature and more like a collaborator — systems that reason, that watch, that listen, that decide. Eight years teaching machines to see the world and occasionally to talk back.
8+
YEARS IN AI
Vision, NLP, agentic systems
50+
SHIPPED MODELS
Prod-grade, not demos
3
ECOSYSTEMS
NVIDIA · HF · OpenRouter
∞
COFFEE UNITS
Required fuel
CHAPTERS · 2016 → NOW
A brief history
of building.
2025
CH · 01

Director of AI

MyBlue.ai
Mar 2025 — Present

Building a video data lake with full agentic capabilities (multimodal RAG, LLM tools, graph-native retrieval).

GenAIAgenticVideoRAGQdrant
2022
CH · 02

Lead AI Architect

SparkCognition
Aug 2022 — 2025

Roadmapped and led SC’s Visual AI Advisor (VAIA). Patented a low-resource face recognition pipeline; built cross-cam re-ID; shipped SDKs.

Computer VisionDeepStreamTensorRTProduct
2023
CH · 03

Visiting Research Scholar (Former)

UC Berkeley
2023

Wildfire detection research: low-latency pipeline across ~1k CalTrans CCTVs; YOLO-based smoke/fire detection.

ResearchCVEdge AI
2020
CH · 04

AI Architect

Integration Wizards
2020 — 2022

Designed high-FPS video analytics on GPUs (DeepStream), modular IRIS pipeline (−75% time-to-deploy).

CUDAPipelinesAnalytics
ARSENAL · 002
The stack,
spoken aloud.

I work primarily in PyTorch and TensorRT, ship agents with LangGraph and DSPy, index reality in Qdrant, stream inference through DeepStream, squeeze models with CUDA, and let Claude do the typing.

GRP · 01
Vision & Edge
DeepStreamTensorRTTritonONNXOpenCVCUDA
V
GRP · 02
Agents & LLMs
LangGraphDSPyOpenRoutervLLMRAGTool Use
A
GRP · 03
Modeling
PyTorchTensorFlowJAXTransformersDiffusionRL
M
GRP · 04
Infra
AWSGCPKubernetesFastAPIQdrantRedis
I
MISSION FILES · 003
Selected
works.
PRJ · 01 · FEATURED
Veronica
VERONICA · AGENT ONLINE
A British-inflected, slightly menacing cognitive agent that represents me on the web. Agentic tool loop over a custom portfolio API; streams thinking, tools, and structured blocks. Lives on Vercel + OpenRouter.
AGENTSSSEFASTAPIOPENROUTER
PRJ · 02
Hermes — Wildfire Detection (DeepStream)
Drone-assisted wildfire detection on Jetson Xavier NX using DeepStream + YOLO; RTSP bridge for Tello.
JETSONDEEPSTREAMYOLO
PRJ · 03
Activity Recognition — TensorRT
3D-ResNet video classification optimized with TensorRT for real-time action recognition.
TENSORRTVIDEO3D-CNN
PRJ · 04
YOLOv4 — OpenCV CUDA DNN
Run YOLOv4 directly in OpenCV’s CUDA DNN for fast CV inference.
OPENCVCUDA
PRJ · 05
CenterFace — DeepStream
CenterFace (ONNX) accelerated on DeepStream 5.1 with custom parsers.
FACEONNXDEEPSTREAM
PRJ · 06
Vector + Advanced AI
Hacking Anki Vector with advanced AI capabilities. Because why not.
ROBOTICSLLM
SIGNALS · 004
Live from
the workshop.
REC · CH 01
FIELD NOTES · 005
Writing.
#001Blaze through your setup with Z by HP Data Science Stack ManagerBLOG →#002What it takes to be an Artificial Intelligence ArchitectBLOG →#003What it takes to be a Deep Learning EngineerBLOG →#004Train models like a pro with NVIDIA TLT 3.0BLOG →#005YOLOv4 with CUDA-powered OpenCV DNNBLOG →#006How an AI organisation can make a difference during the pandemicBLOG →#007Computer Vision: A Step Closer to SkynetBLOG →
HONOURS · 006
Notable.
★
PATENT
Patent: Face Image Matching based on Feature Comparison
★
AMBASSADOR OF THE MONTH — Z BY HP
Ambassador of the Month — Z by HP
★
NVIDIA
Jetson AI Ambassador · Deep Learning Institute instructor
★
SPEAKING
GTC · NeurIPS meetups · AI Engineer Summit · 20+ conferences
★
CREATOR
YouTube channel on applied AI
ESTABLISHING TRANSMISSION

Let's build
something
outrageous.

OPEN CHANNEL
EMAIL · [email protected]GITHUB · /kingardorLINKEDIN · /akashjamesMEDIUM · akash-james.medium.comTWITTER · @king_ardor