00.00.00
Aditya Shaji

Aditya Shaji

Software developer, Mumbai

Command Palette

Search for a command to run...

I build agent systems, and the harnesses that check them.

Read the resume
Send an email

Here are my socials

Experiences

Low End LLM
Low End LLM
Founder
Jul 2026 - Present
Remote

4 GB

Target RAM

Per turn

Safety floor

Q4_K_M

Quantization

In progress

Status

  • •Building an adaptive runtime that runs quantized LLMs locally on 4GB-RAM laptops, for developers without high-end GPUs who want a real, local taste of tools like Claude Code and Codex
  • •Shipped real-time RAM, CPU and battery profiling plus a per-turn safety floor that halts inference before memory exhaustion instead of letting the OS swap
  • •Integrating quantized model loading (GGUF via llama.cpp, Q4_K_M, mmap-based weights) behind a pre-flight capacity check that refuses a model the machine cannot hold
  • •Treating the memory budget as an explicit, checked constraint rather than something discovered by crashing - the same discipline the agent repos apply to context and state
Pythonllama.cppGGUFQ4_K_MmmapMemory budgeting
Redis State Explorer
Redis State Explorer
Creator & Maintainer
Jun 2026 - Present
Remote

2,000+

Installs

200+

Commands

<1 s

Update latency

MIT

License

  • •A VS Code extension that connects straight to Redis over the wire with ioredis - no backend, no extra service, sub-second updates
  • •Save-diff timeline: snapshot Redis state before and after a file save, then show exactly which keys the save caused to change
  • •200+ command terminal with syntax highlighting, a live key explorer for every Redis data type, and a code detective that maps keys back to the lines that reference them
  • •2,000+ installs on the VS Code Marketplace, MIT licensed, actively maintained
  • •The save-diff idea is the one that keeps coming back: record the before and after, then let the evidence - not the tool - say what happened
TypeScriptRedisioredisVS Code APITLS

Projects

Play Video

AnyBridge preview

AnyBridge

In progress

An MCP server that turns a model's plan into a human-approved, engine-verified workflow - then replays it with no model calls.

View Project

Play Video

AgentTrace preview

AgentTrace

Shipped

When an agent's checkout changes mid-flow, verify the buyer's original mandate before attempting recovery - or abort.

Razorpay
View Project

Play Video

Redis Live preview

Redis Live

Shipped

See exactly which Redis keys your code changed, the moment you save a file. 2,000+ installs on the VS Code Marketplace.

View Project

Play Video

Low End LLM preview

Low End LLM

In progress

An adaptive runtime that runs quantized LLMs on 4GB-RAM laptops, with a per-turn safety floor instead of a crash.

llama.cpp
GGUF
Q4_K_M
View Project

GitHub Activity

Loading GitHub contribution activity

Calendar heatmap showing daily GitHub contribution counts for ps-aditya over the last year. Scroll horizontally to inspect all weeks.

JanFebMarAprMayJunJulAugSepOctNovDec
Less active
More active

Skills & Technologies

MCP
Tool calling
Agent loops
Guardrails
Eval harnesses
Human approval
Structured output
Replay
TypeScript
Python
Java
Go
Node
Playwright
Redis
Docker
VS Code API
Claude
Gemini
llama.cpp
scikit-learn
pandas
Git
Actions
Azure ML
Linux

"The model saying it worked is not evidence.The evidence is evidence."

WHAT I BUILD AGAINST