Скачать или смотреть Giga Chkhikvadze: Desearch, Real-Time Search API, Bittensor Subnet 22, AI Agents, LLMs | Ep. 67

In this episode we are joined by Giga of Desearch, operators of Bittensor Subnet 22. Giga (Founder) shares his background in software development and AI agents, and how Desearch acts as a real-time search API for developers building AI applications. He discusses real-time data integration for LLMs, supported platforms like X/Twitter and Reddit, miner incentives for stability and relevance, collaborations with other subnets, the Rizzy lead generation agent, validation mechanisms, and future expansions.

In this episode, you’ll learn
Giga’s 20+ years in tech, from software development companies to building KYC and invoicing products, and pivoting to open-source AI agents
The origins of Desearch as a solution for real-time data access in AI, addressing LLM "blindness" without current info from social media and the web
Key features like low-latency searches (milliseconds to seconds), fast vs. detailed models, and customizable sources for queries
Popular use cases, including crypto tracking, sports predictions, and forecasting tools built on Desearch’s X/Twitter API
How Bittensor’s incentive mechanics drive miner competition for high-relevance results, stability through distributed queries, and benchmarks against centralized services
Collaborations with subnets like 23, 51, and 64, plus the Rizzy agent for lead generation on X
Revenue models from pay-as-you-go APIs, enterprise deals, and subscription-based apps like Rizzy, with plans to expand sources (e.g., LinkedIn, TikTok)
Giga’s first impressions of Bittensor as a revolutionary economy blending competition and incentives
Advice for subnet developers and miners on focusing on strong areas, building robust incentive mechanisms, and maintaining high-quality, transparent code

Chapters
0:00 - Highlights
1:22 - Introduction
2:11 - What is Desearch? Why Build a Real-Time API?
3:38 - Supported Platforms
4:28 - Products Built Using Desearch API
5:57 - Latency and Performance
6:55 - Fast vs. Detailed Models
7:56 - Specifying Search Sources
8:39 - Most Popular Sources
10:17 - Biggest Selling Points: Cost, Stability, and Relevance
12:50 - Achieving Stability Through Miners
14:15 - Query Distribution Process
15:02 - Relevancy Benchmarks and Comparisons
17:26 - Experience from Building AI Agents
20:01 - Pivoting to Search Infrastructure
21:06 - Giga's Background in Tech and Startups
24:09 - First Impressions of Bittensor
25:27 - Collaborations with Other Subnets
27:02 - Rizzy Agent: Purpose and Creation
29:01 - Future Plans for Rizzy and New Features
30:23 - Profitability: API vs. Applications Like Rizzy
31:49 - Expanding Data Sources
32:37 - Validation Process for Miners
34:34 - Rate Limits and 80% Burn Impact
36:32 - Ownership Structure and Co-Founders
37:57 - Vision for Desearch's Future
39:45 - Advice for Subnet Developers and Miners