AI Agents, Language Models, Systems

Aadit Suryawanshi

I'm currently building AI agents capable of planning, reasoning, and using tools to solve real-world problems. Alongside agent systems, I work on memory architectures, retrieval pipelines, and language model training to better understand and build reliable AI systems end-to-end.

GitHub

Projects

LiteGPT

A clean decoder-only Transformer language model (~25M parameters) trained from scratch on a single NVIDIA A5000 GPU.

Recall-OS

RecallOS is an AI-native enterprise knowledge operating system that allows organizations to ingest, organize, search and reason over every piece of company knowledge.

SmolLM-135M_Med

End-to-end pipeline for continued pretraining, supervised fine-tuning, and evaluation of SmolLM-135M on medical datasets.

Xcal

Realtime Excalidraw

Current rabbit holes

  1. Experimenting with how AI systems can remember, retrieve, and reason over knowledge across conversations and documents.
  2. Building my own memory architecture.
  3. Fine-tuning open-source LLMs for medical QA
  4. Training GPT-style language models from scratch in PyTorch

Featured article

Building a 25M parameter GPT from scratch

A practical write-up on architecture choices, data mixtures, tokenizers, training issues, and what I learned after training LiteGPT on FineWeb and TinyStories.

Read article