Skip to content
View sourabhbgp's full-sized avatar
πŸ’»
Coding ...
πŸ’»
Coding ...

Block or report sourabhbgp

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
sourabhbgp/README.md

Hey, I'm Sourabh πŸ‘‹

7 years shipping software. Now building AI agents that do real work.

About Me

  • 7+ years building production software across startups and enterprise SaaS
  • Currently diving deep into open source and the AI agent ecosystem
  • Based in Bengaluru, India

Tech Stack

TypeScript React Next.js Node.js PostgreSQL Docker Puppeteer Vercel AI SDK LLM Orchestration


Projects

Closed Source

  • chromext.app β€” Chrome extension for flash-sale automation. 5,000+ active users
  • lootkar.com β€” Multi-agent shopping assistant for Indian e-commerce (in development)

Open Source

  • sotto β€” Voice input for Claude Code via local whisper.cpp
  • ecom-extract β€” Extract structured product data from any e-commerce URL
  • dikto β€” Voice-to-text for macOS. Local transcription via Whisper + Parakeet. No cloud. (Rust, SwiftUI)

What I'm Working On

Twitter Agent (coming soon β€” open source)

CLI-first autonomous Twitter engagement agent. You define workflows (follower-growth, hashtag-niche, or custom), and the agent runs on autopilot β€” fetching your feed from three sources (mentions, timeline, keyword discovery), having Claude analyze each tweet, and taking actions (reply, like, quote, retweet, follow) within configurable rate limits.

What makes it interesting:

  • Workflow system β€” Each workflow has its own strategy prompt, action biases (heavy/moderate/light per action type), feed filters, watch accounts, and isolated memory
  • Graph memory (Neo4j) β€” Tracks per-handle relationship strength, reply-back rates, topic history, and interaction patterns. This context is injected into Claude's system prompt so it knows "I've talked to this person 12 times about AI agents, they reply back 40% of the time"
  • Scheduling β€” Cron-based job scheduler with platform-native backends (launchd on macOS, systemd on Linux, in-process daemon for Docker). Set it and forget it
  • Safety guardrails β€” Spam pre-filtering, blocked account lists, session/daily action caps, global safety state shared across workflows
  • Two modes β€” Auto (Claude decides everything) and manual (interactive approve/edit/skip loop per tweet)

Built with Bun, TypeScript, Commander.js, Anthropic SDK, and rettiwt-api. Includes a read-only Next.js dashboard for monitoring.


Connect

Twitter LinkedIn

Pinned Loading

  1. sotto sotto Public

    sotto voce β€” voice input for Claude Code via local whisper.cpp

    TypeScript

  2. diktoapp/dikto diktoapp/dikto Public

    Voice-to-text for macOS. Records speech, transcribes locally. Whisper + Parakeet. No cloud.

    Rust 6 1

  3. ecom-extract ecom-extract Public

    Extract product prices, page content, and bank offers from Indian e-commerce sites.

    TypeScript