Shivam Tiwari

AI Specialist | Agentic AI & MLOps. 🤖

Deep diving into autonomous agent architectures and hybrid neural networks.

Building the future of AI-integrated system

[Shivam Tiwari @ncdevshiv]

@elonmusk Update on the MT5 project:

I’ve pivoted from the main framework to focusing on the EA for now. I’m currently working 9–9 on a $240/month job, so I haven’t been able to dedicate as much time to the project as I want to.

If you’d like to support the development and help me work on it more seriously, you can support here:
https://t.co/8z5vOMkO4p note i do not even know how i can get support i live in india pls tell me best method to get support
Now for the update 👀

While building a Forex EA, I accidentally created a god-tier Gold EA.
If you had been using this EA over the past 5 months, $100 could’ve turned into $3,700.

More updates soon 🚀

[Shivam Tiwari @ncdevshiv]

NCDev Codex! A fork of openai codex,

I have made several updates to the Codex CLI and now need further modifications.

Currently, Codex CLI only supports the OpenAI provider and OpenAI models, and users must log in with a paid/Plus-level account to use it. I have already started modifying this behavior.

Your task is to fully remove these authentication restrictions and provider limitations, and convert Codex CLI into a model-agnostic and provider-agnostic system.

Goals

Remove OpenAI-only restriction
The CLI must no longer depend solely on OpenAI authentication.
It should work with multiple providers.

Introduce a /model command
When the user runs /model, a popup interface should open.
First-time provider selection
On the first run, the popup should show a list of supported providers:
OpenAI
OpenRouter
Users can select a provider from the list.

Provider authentication options
OpenAI
Option 1: Provide API key
Option 2: Login (existing method)
OpenRouter
Provide API key

Dynamic model fetching
The system must NOT contain any hardcoded model list.
Models should be fetched dynamically from the provider API.

Model selection UI
Since some providers offer 100+ models, implement:
Smart search
Filtering
Fast navigation

Persistence
Save:
Last used provider
Last selected model
API keys (securely)
On future runs, /model should:
First load models from the last used provider.

Switching providers
Inside the model popup:
Press Ctrl + A to open the provider selection list.
Adding custom providers
In the provider popup:
Press Ctrl + P to open a new popup for adding a custom provider.
Required inputs:
Provider name
Base URL
API key
The system should then:
Fetch and display available models from that provider.
General requirements
The implementation must be end-to-end complete.
Think through the full recursive flow of the system (UI, persistence, provider abstraction, model fetching, etc.).
Ensure compatibility with future providers.
Follow rules.md strictly and do not violate any defined rules.

The final result should be a fully provider-agnostic, model-agnostic Codex CLI with dynamic model discovery, provider management, and persistent configuration.

https://t.co/IAGkRdmW4n

#AI #Coding #OpenSource #codex

[Shivam Tiwari @ncdevshiv]

MQL5 is great for charts, but it’s a bottleneck for high-performance systems. If you're building a quant terminal, you need a direct line to the engine.

Meet Titan Bridge: an open-source, high-speed bridge designed to connect MetaTrader 5 to modern, low-latency environments.

This is the first beta version. You may encounter bugs or issues—please report them, or fix them directly and request a merge.

Please support me: https://t.co/g8i4h0MaNa

💎 The Performance Edge:
• Ultra-low latency: optimized for execution speed
• Rust-powered: harness the safety and speed of systems-level code
• Real-time streaming: zero-lag market data for your models
• Async-ready: built for modern, concurrent trading architectures

Stop settling for "fast enough." Take full control of your execution.

Give it a ⭐ and contribute: https://t.co/Lk33vdFnzK

#QuantTrading #RustLang #AlgorithmicTrading #MT5 #SystemsEngineering @ncdevshiv

[Shivam Tiwari @ncdevshiv]

How do you enforce agents to follow a certain set of rules?
Any suggestions? It feels like I'm working with a set of liars, lazy cheaters who only try to fool me with fake promises and claims, and keep breaking the rules over and over.🥺

#AIAgents #PromptEngineering #AIAlignment #AIGaslighting #LLMFail #FrustratedDev #RuleBreakingAI #AIdev #DevStruggles

[Shivam Tiwari @ncdevshiv]

https://t.co/SQkrCF2wO0

[Shivam Tiwari @ncdevshiv]

Very soon I will be 🚀 announcing the open-source release of the Titan DMA Bridge—an ultra-low-latency MetaTrader 5 (MT5) to Rust bridge designed for the next generation of HFT systems. 🦀📈

We’ve pushed the boundaries of MQL5 to achieve sub-10μs tick processing. If you are building high-frequency quant terminals, this is your new foundation.

⚡ Why it’s different:
True Zero-Copy: Powered by Eclipse iceoryx shared memory. No more ZeroMQ overhead. Data moves at the speed of RAM.

Massive Throughput: Engineered to handle 200,000+ ticks/sec across 2,000+ symbols simultaneously.

Agnostic Architecture: Fully decoupled from your DB or frontend. Plug it into DuckDB, SpacetimeDB, or your own custom stack via a raw binary Protobuf interface.

Smart Normalization: Automatically maps "GOLD" (XM) or "XAUUSD" (Vantage) to a single Canonical ID (CID) for unified analysis.

🛡️ Production-Grade Reliability:
Hardened Safety: Integrated circuit breakers and max-slippage guards to protect your capital during data micro-bursts.

Stealth Execution: Supports virtual SL/TP and Iceberg orders managed entirely in Rust—stop hunting ends today.

Hermetic & Portable: One setup script handles all dependencies (RouDi, static memory segments, and DLLs). No "it works on my machine" errors.

This isn't just a bridge; it's a Direct Memory Access (DMA) gateway that replaces the slow native Python API with a high-performance Rust execution slave.

The repo will include a full Handover Manifest, a complete Error Registry (no more silent failures!), and a system wiki to get you up and running in minutes.

Now finding edge cases, testing, and benchmarking it under various loads—and also deciding which licence to use when open-sourcing this.

#RustLang #HFT #TradingTech #OpenSource #MetaTrader5 #QuantFinance #SystemsEngineering

[Shivam Tiwari @ncdevshiv]

Very soon I will be 🚀 announcing the open-source release of the Titan DMA Bridge — an ultra-low-latency MetaTrader 5 (MT5) to Rust bridge designed for the next generation of HFT systems. 🦀📈 Just thinking about which licence to choose.

#RustLang #HFT #TradingTech #OpenSource #MetaTrader5 #QuantFinance #SystemsEngineering

[Shivam Tiwari @ncdevshiv]

Hi, I'm building a fully AI-native trading terminal and want to use a chart library, but Lightweight Charts isn't sufficient. Can anyone suggest the best chart library for this? I'm also open to donations- https://t.co/g8i4h0MaNa

#TradingTerminal #FintechDev #ChartLibrary #TradingCharts #FinancialCharts #JavaScriptCharts #TradingViewAlternative #AdvancedCharts #BuildInPublic #BuyMeACoffee

[Shivam Tiwari @ncdevshiv]

Old first-gen model trades, please note: These are very first-gen and were mostly guesses, meaning even AI can have beginner's luck.😅

#AIFirstGen #BeginnerLuck #OldAIModels #AITrading #EvenAIgetsLucky #TradingAI #AILuck #QuantTrading

[Shivam Tiwari @ncdevshiv]

🚀 MT5 Traders Poll!

Who is using MetaTrader 5 for trading?

And are you able to code/setup .MQ5 Expert Advisors (EAs) yourself?

Vote below 👇
#MT5 #MQL5 #MT5EA #ForexTrading #AlgoTrading #ExpertAdvisor

[Shivam Tiwari @ncdevshiv]

@duckdb Please note that even after this, I am still using DuckDB since it remains one of the best databases, and I have paired it with SpacetimeDB to mitigate multiple issues.😊

[Shivam Tiwari @ncdevshiv]

@duckdb and also i have added a back sync to the system where it auto detect for how much time it was either disconnected or offline and write back to it i noticed on rare occation i am facing curruption and i had to add a manual replication and time driven backup.

[Shivam Tiwari @ncdevshiv]

@duckdb Hi, I may be wrong, but when I tried to write multiple markets tick and ohlcv data independently to the DB, I noticed writes are serialized and not parallel.

[Shivam Tiwari @ncdevshiv]

Stuck on charts… Lightweight Charts just isn’t cutting it for an AI-native terminal.

Need something that handles:
• Super smooth 1000+ pair rendering
• Custom AI overlays
• Scripting support

Drop your best chart lib recommendations 👇

#Dev #TradingTerminal #ai

[Shivam Tiwari @ncdevshiv]

There were two directories: one named ai-ag and the other ai-ag-py.

I prompted Windsurf to delete the ai-ag directory, find where it is linked, and stop that process.

Even after so much development, this dumb AI deleted ai-ag-py instead of what I asked.

[Shivam Tiwari @ncdevshiv]

After a long development process, I'm facing these issues. Can anyone help?

1. What is the best way to handle over 200,000+ ticks across 1000+ market pairs?
Currently I'm using ZeroMQ.

Also, the main issue is the frontend. Can anyone suggest the best stack for this?