Your AI workforce across every app, every task, every day. Core infrastructure powering the global agent ecosystem. Builder of @HeySorinAI
Global
Joined July 2021
- Tweets3.5K
- Following117
- Followers584K
- Likes3.3K
Pinned Tweet
What institutional trading desks have always had, now in the hands of every investor.
Powered by @SaharaAI
Sorin is live.
Your personal AI agent for global digital markets.
Get started at HeySorin.AI
This is what Agent-as-a-Service is built for. 馃
Excited to see @DeAgentAI building with @SaharaAI & @HeySorinAI. 馃攩
This is what we mean when we say AI should go beyond answering questions.
@HeySorinAI is built to understand the market, turn ideas into strategies, and take action.
From intelligence to execution - this is the kind of agent we鈥檙e building at Sahara AI. 馃攩
The historical data you can access shapes what you can learn from a quant strategy. A wider span lets you revisit the same rules in different periods. Some signals keep showing up; others only appeared in a particular stretch of prices. Both findings tell you what to test next.
With more history to choose from, you can be deliberate about the dates. Keep the strategy settings fixed and move the window. When its behavior changes, you have a specific period to investigate. That gives you more context before you revise the rules or move toward live trading.
We've expanded the selectable history in Sorin backtests. Custom Range can now reach back up to three years, with up to six months in a single backtest. Choose a quant strategy, open Backtest, and select the period you want to study.
AI products will win on cost per useful answer.
Diffusion can draft text and code in blocks; routing can send easy tasks to small models. Both may cut wait time and cost if quality holds. That leaves room for profit at scale.
Here鈥檚 how both work and where they can fail. 馃У馃憞
Most models write one token at a time. Diffusion can draft a block of text at once, then refine it.
That could speed up code edits or live chat. Yet each pass uses compute, and fewer passes can hurt quality. The fair test is the full time and cost to get a good answer.
Introducing Sorin's upgraded Trading Copilot, with Hyperliquid integration.
Select AI Analysis and describe the trade you want. Sorin explains the reasoning and recommended settings, then fills the order card after you confirm the parameters. 馃У馃憞
For example: with 21.25 USDC available, this ETH short order is sized at 212.39 USDC using 10x leverage and a 100% allocation.
You can see how the position is sized before accepting it.
Agentic finance is moving from concept to real-world execution.
Great to see @CambrianNetwork pushing this space forward alongside @HeySorinAI.
As agents gain better access to intelligence, liquidity, and execution, we鈥檙e getting closer to a financial ecosystem where AI can not only understand markets - but act on them.
The agentic economy is being built now. 馃
Our co-founder @xiangrenNLP is at @COLM_conf in SF this week 馃摲
As LLMs become agents that act on our behalf, the big question is: can we actually trust and verify them?
He'll be sharing our work on securing and auditing LLMs, from vulnerability detection to provenance to agent safeguards. If you're around, say hi 馃摲
馃帀 Excited to team up with @DGrid_AI to empower the next wave of AI agent builders on Sahara AI.
By combining DGrid鈥檚 Smart Router and Decentralized Model & Agent Marketplace with Sahara鈥檚 Tools-as-a-Service, developers can access and compose high-quality models, agents, and AI tools through a more open, efficient, and decentralized infrastructure.
Build once. Route smarter. Access more AI capabilities. 馃殌
About DGrid
DGrid AI is a decentralized AI smart network dedicated to building an open, transparent, and community-driven decentralized AI network. By establishing a unified smart marketplace and scheduling layer, it provides developers and users with low-cost, verifiable, and trustless access to models and AI Agents.
An AI research result can be right and still have no credit trail.
The final answer may show proof and sources, yet hide the agents, data, tools, and people behind it. If the steps that changed the outcome are lost, credit and pay become guesswork.
Read on for the workflow. 馃У馃憞
One agent finds the key paper and another runs the test.
A human spots a flaw in the data and sends the work back. That call changes the result. The final answer can cite the paper and show the proof while leaving the human correction out of the credit trail.
Log the work as it goes. Note the model that found a lead, the data that shaped the test, and the human who caught a flaw.
Keep each step tied to that result so credit and pay can follow it. @SaharaAI is building attribution and automatic value distribution for agent workflows.