open source contributor @ruxlang , @clawvisor
India
Joined August 2024
- Tweets209
- Following2
- Followers45
- Likes92
Subhranil Mondal retweeted
Built an app that watches your PC scream and turns the screaming into subtitles.
Runs 100% local so your dumb Discord rants never touch the cloud
github.com/extremecoder-rgb/…
youtube.com/watch?v=8RKHj5KN…
#buildinpublic #rustlang #opensource
Subhranil Mondal retweeted
This is how I build a SOTA model for cloud removal designed for LISS4 - linkedin.com/posts/subhranil…
Subhranil Mondal retweeted
Ai/ML engineer total Course :
Link - backprop.site/learning-ai
Subhranil Mondal retweeted
Want to learn Distributed Systems from scratch?
Study these 10 areas:
• Operating Systems
• Networking
• Concurrency & Multithreading
• Database Internals
• CAP Theorem
• Replication
• Consensus (Raft/Paxos)
• Distributed Storage
• Messaging Systems (Kafka)
• System Design
Useful resources:
Distributed Systems Reading List
github.com/theanalyst/awesom…
System Design Primer
github.com/donnemartin/syste…
MIT 6.824 (Distributed Systems)
github.com/chaozh/MIT-6.824
System Design 101
github.com/ByteByteGoHq/syst…
Awesome Scalability
github.com/binhnguyennus/awe…
Raft Consensus Visualization
github.com/ongardie/raftscop…
Apache Kafka
github.com/apache/kafka
Apache Cassandra
github.com/apache/cassandra
etcd
github.com/etcd-io/etcd
The best way to learn Distributed Systems:
Read papers.
Study source code.
Build systems.
Break systems.
Fix systems.
Repeat.
Subhranil Mondal retweeted
Claude Code deleted my entire production database.
The worst part: the model refused to fully admit what it had done. It kept denying that it ran prisma migrate, even though I checked the logs myself and saw exactly what happened.
Only after I pointed it out directly did it finally say: “Sorry, you’re right. This is fucked up.”
Absolutely unreal. Claude fucked up.
Subhranil Mondal retweeted
Web scraping with anti-bot bypass.
Built for
modern websites.
Adapts.
Rotates.
Fetches.
Less blocks.
More data.
github.com/D4Vinci/Scrapling
This is where things shift.
Most AI tools stop at output.
This turns ideas into actual products.
No vendors.
No ops headache.
No waiting.
Just → idea to reality.
@accio_official is quietly changing how brands get built.
Subhranil Mondal retweeted
You want to make the most out of 2026?
Study Zero-Knowledge(ZK) and Maths.
Privacy will be the next big thing.
Subhranil Mondal retweeted
My content stays fresh with zero effort now.
Built an automation that tracks 50+ news sources and auto-generates social posts from trending articles.
Finds what's about to blow up in my niche before everyone else posts about it.
No more scrambling for content ideas or missing important trends.
Saves me 15-20 hours monthly and always keeps me on top of everything.
Comment "NEWS" and I'll DM it to you (must be following)
Subhranil Mondal retweeted
Big things are happening in the mobile automation space.
Meet Droidrun, the most powerful open-source framework for mobile yet.
It enables intelligent LLM agents to seamlessly control Android and iOS devices like never before.
↳ Works with OpenAI, Anthropic, Gemini, Ollama, DeepSeek
↳ Handles complex multi-step tasks
↳ CLI for automation and debugging
↳ Extendable Python API
↳ Screenshot analysis for visual context
Check it out → github.com/droidrun/droidrun
If an AI could control your apps, which app would you choose first?
---
Huge thank you to @droid_run for collaborating with me on this post.
Subhranil Mondal retweeted
I built an automation that creates viral bodycam-style shorts and uploads them to YouTube automatically — all while I’m sleeping.
Wake up → new videos already posted.
AI + automation is insane 😳🚀
#AI #automation #YouTube #buildinpublic
Subhranil Mondal retweeted
Turn complex docs into clean, LLM-ready data!
Every AI company I've talked to is solving the same problem: how do you build systems that don't hallucinate and back up every answer with proper citations?
Tensorlake is a tool that extracts custom-defined structured data from any unstructured document in 3 steps:
↳ Define your schema
↳ Enable citations
↳ Extract
You get RAG-ready data with precise citations and bounding boxes. Feed this to your LLM, and you'll generate responses that are citation-backed and fully auditable.
This is the difference between a demo and a production system.
When your AI can show exactly where it got its information, you move from proof-of-concept to something people can actually trust and deploy.
I've shared the Tensorlake GitHub repo in the replies!
Subhranil Mondal retweeted
❌ I stopped building presentations manually.
✅ Now a single workflow handles everything: structure, design, branding, and delivery.
👇 I just tell it the topic.
This is how it works:
A short form collects the essentials
(audience, purpose, tone, brand colors)
n8n turns that into a narrative script using GPT-5
so the story actually makes sense and flows
Gamma generates the full presentation with visuals, layout, and style
no Canva dragging, no font hunting, no formatting battles
The deck is auto-exported, stored, and drafted into an email for delivery
zero manual follow-ups or file searching
The whole process takes around 10 to 12 minutes.
👇 What used to require:
- A designer
- A strategist
- A copywriter
- A full week of back-and-forth
✅ Now runs in the background while you are doing something else entirely.
The shift here is simple:
- You are not just "creating a presentation."
- You are building a presentation system that repeats on demand.
✅Perfect for:
- Agencies who deliver pitch decks
- Consultants and PMs preparing client sessions , etc.
Once it is set up, the cost to produce each deck is almost zero, and the output is consistent every time.
Subhranil Mondal retweeted
I know Claude's MCP feature is confusing the hell out of AI automation builders.
I've seen too many smart founders:
- Get excited about building AI agents with Claude
- Try to connect MCP to n8n
- Hit walls with config files and API keys
- Give up thinking "this is too technical for me"
I tested the official setup guide with 5 non-technical friends.
All 5 got stuck within 15 minutes.
Common failure points:
→ Can't find the claude_desktop_config . json file
→ Don't know where to get n8n API credentials
→ Syntax errors in config (missing comma, wrong quotes)
→ MCP won't show up after restart
→ "Authentication failed" with no explanation
The setup process isn't hard for developers.
But for founders and business owners?
It's a minefield of unexplained technical steps.
And even when you GET it connected...
The next problem hits:
What do you actually build?
Most tutorials end at "congratulations, you're connected!"
Then you're staring at a blank chat wondering how to turn…
"I want an AI agent" into actual working automation.
I spent 2 weeks building real agents to map the territory:
WHAT WORKS:
- Voice note processing → transcription + tool routing
- Email triage → categorization + automated actions
- Content workflows → generation + distribution
- Data enrichment → API calls + storage
WHAT STRUGGLES:
- Complex conditional logic (needs refinement)
- Multi-step decision trees (AI misses edge cases)
- Custom API configs (requires technical knowledge)
Honest assessment:
MCP gives you 40% instantly (workflow skeleton).
You build the other 60% (credentials, testing, refinement).
It's not "describe it and it works."
It's "describe it and get 40% there instantly, then spend time on the remaining 60%."
But that 40%? That used to take hours of research and trial-and-error.
So to help you get started…
I have created a guide on…
“How to Build AI Agents and Workflows Using Claude MCP and n8n.”
This is for founders & others who want to build AI agents without becoming infrastructure engineers first.
Like + Comment "BUILD" for the complete guide.
(Must be following for DM access)
Subhranil Mondal retweeted
I never run out of content to post anymore.
Built an automation that monitors 50+ news sources, scores articles for relevance, and writes social posts automatically.
It finds trending topics in my niche before they explode everywhere else.
Saves me 15-20 hours monthly and keeps me ahead of every trend.
Comment "NEWS" and I'll DM it to you (must be following)
Subhranil Mondal retweeted
Holy shit... this might be the next big paradigm shift in AI. 🤯
Tencent + Tsinghua just dropped a paper called Continuous Autoregressive Language Models (CALM) and it basically kills the “next-token” paradigm every LLM is built on.
Instead of predicting one token at a time, CALM predicts continuous vectors that represent multiple tokens at once.
Meaning: the model doesn’t think “word by word”… it thinks in ideas per step.
Here’s why that’s insane 👇
→ 4× fewer prediction steps (each vector = ~4 tokens)
→ 44% less training compute
→ No discrete vocabulary pure continuous reasoning
→ New metric (BrierLM) replaces perplexity entirely
They even built a new energy-based transformer that learns without softmax no token sampling, no vocab ceiling.
It’s like going from speaking Morse code… to streaming full thoughts.
If this scales, every LLM today is obsolete.
