Skip to content
View airawatraj's full-sized avatar
🎯
Focusing
🎯
Focusing
4 followers · 5 following

Achievements

Achievement: YOLO

Achievements

Achievement: YOLO

Block or report airawatraj

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
airawatraj/README.md

technologist | experimenter | seeker

"the future belongs to those who understand at a very deep level how to combine their unique expertise with what algorithms do best" — Pedro Domingos, The Master Algorithm


Portfolio

Independent Research & Projects.
  • Cogni-Brain-2: Achieved 218.85 TPS and a perfect 100/100 tool-use score running Qwen 3.6-35B on NVIDIA DGX Spark via the Atlas engine. Explored the trade-offs between ultra-fast agentic execution and the deep-reasoning stability of larger parameters.
  • Cogni-Brain: Designed and optimised a secure, sovereign, production-grade agentic AI system on NVIDIA DGX Spark using NemoHermes, OpenShell, and kernel-level sandboxing; advanced community benchmarks.
  • SageGPT-7.5M (NVIDIA DGX only): SLM trained from scratch on ~140M pure Sanskrit tokens on NVIDIA DGX Spark. 6 Layer, 8 Attn Head, 256 embed, 1024 context.
  • SageGPT-7M (MLX only): SLM trained from scratch on ~57M Sanskrit tokens on Apple Silicon. 4 Layer, 8 Attn Head, 256 embed, 256 context.
  • Cogni.chat: Local-First Multimodal AI Agent - orchestrating Cogni Still, Cogni Breath, Cogni Ground, Cogni Weaver, and Cogni Sign into one cohesive intelligence designed to support the human mind, body, and spirit.
  • Fiduciary-Ops-Agent: An autonomous enterprise governance agent utilising a strict Check-then-Act protocol via Gemini 2.5 Flash Lite; enforces real-time fiduciary risk-alignment (CLV vs. Refund) using tool-first orchestration.

"I have no special talent. I am only passionately curious" - Albert Einstein

Pinned Loading

  1. dgx-spark-qwen-super-agent dgx-spark-qwen-super-agent Public

    Cogni-Brain-2: Qwen 3.6-35B on DGX Spark via Atlas. Hitting 218.85 tok/s (9x speedup) and a perfect 100/100 Tool-Eval with local NVFP4 acceleration.

    Python

  2. dgx-spark-nemotron-super-agent dgx-spark-nemotron-super-agent Public

    Nemotron-3-Super-120B on DGX Spark (~24 tok/s): vLLM config, hardened agentic stack, and benchmark methodology

    Python 4 3

  3. sage-gpt sage-gpt Public

    SageGPT (7.5M param SLM): A Transformer trained from scratch on ~140M pure Sanskrit tokens on NVIDIA DGX Spark. 6 Layer, 8 Attn Head, 256 embed, 1024 context, ~8K vocab.

    Python

  4. fiduciary-ops-agent fiduciary-ops-agent Public

    An autonomous Fiduciary Agent powered by Gemini Flash Lite that enforces enterprise risk governance (CLV vs Refund) using strict tool-first protocols

    Jupyter Notebook

  5. sage-gpt-mlx sage-gpt-mlx Public

    SageGPT 7.25M param SLM trained from scratch on 56.89M Sanskrit tokens using Apple MLX. 4-layer decoder-only Transformer with 8K vocabulary for Apple Silicon inference.

    Python