Shweta Mishra
320 posts

Shweta Mishra
@ShwetaTechNova
AI & Data Scientist | ML | Data Analysis Founder @TechNovaWorld Daily AI insights • Learn. Innovate. Grow. 🚀 Human Vision + AI = Reality 🚀
Moradabad India Katılım Kasım 2019
12 Takip Edilen11 Takipçiler

Building TokenMizer in public — an AI memory layer with graph-based context compression.
pip install coming soon. MCP server live now.
If you're building LLM agents and fighting context bloat, I'd love your feedback github.com/Shweta-Mishra-…
#OpenSource
English

Mem0. Zep. Letta.
All solving "AI memory" differently. None of them give you a visual graph explorer + CLI + MCP server in one package.
That's the gap TokenMizer fills. Built it because I needed it myself.
5. CTA/community hook
#Opensource #github #TokenMizer
English

Spent the last month finding and fixing 9 production bugs in my own tool before launch — race conditions, silent crashes, broken regex logic.
Boring work. But it's why TokenMizer's test suite is solid now.
Ship quality > ship fast.
#OpenSource #python #TokenMizer
English

Every AI agent I built kept forgetting context between sessions.
So I built the fix myself: TokenMizer — persistent graph memory for LLM apps, with a D3.js explorer so you can literally see what your agent remembers.
#OpenSource #llm #TokenMizer #graphmemory
English

Your LLM API bill is 90% wasted tokens.
Built TokenMizer to fix that — a graph-memory proxy that compresses context before it hits the model. SQLite-backed, MCP-native, works with any LLM stack.
Free tier live. Try it 👇
github.com/Shweta-Mishra-…
English



































