Skip to content
View rohith4444's full-sized avatar
πŸ§‘β€πŸ’»
Determined.
πŸ§‘β€πŸ’»
Determined.

Block or report rohith4444

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. Markdown is supported. 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
rohith4444/README.md

πŸ‘‹ Hey, I’m Rohith β€” AI Engineer & Builder

I build voice-first AI products and multi-agent systems. With 5+ years of experience in AI/ML engineering, I have experience in Conversational AI, Voice AI, and Multi-Agent Systems.

My focus is on building intelligent systems to change the way we communicate with computers and create more personalized AI experiences that adapt to individual users and business needs.


πŸ› οΈ What I Do

  • Build voice-first AI products with real-time streaming (WebSocket, STT/TTS)
  • Design multi-agent systems with Google ADK and LangChain
  • Ship end-to-end AI applications from prototype to production (GCP, Cloud Run)
  • Develop Chrome extensions with AI integration
  • Create RAG pipelines and LLM-powered applications

πŸ† Achievements

πŸ₯‡ AI Hackathon Winner - Corp-Agent Vibe Coding (San Francisco, Aug 16, 2025)

The Challenge: In a 4-hour sprint, our team tackled a real sales pain point β€” companies receive tons of inbound calls daily, with sales reps wasting valuable time reviewing call transcripts in CRMs, often spending ~5 minutes per contact just to figure out if a lead is worth pursuing.

Our Solution: Built an AI-powered sales-call filtering system (cheekily named "Legit or Sh*t" by the team πŸ˜…) that automates the entire lead qualification process using a two-layer AI system that transcribes calls, adds real-time company context, and intelligently classifies them as potential leads, spam, or uncertain.

Technical Implementation:

  • Workflow 1 (Automatic): Triggered in real-time during live calls, instantly labeling leads before they enter the CRM system
  • Workflow 2 (Manual): CRM-integrated batch processing to label and filter existing contacts at scale

Impact: Sales teams can save 1000+ hours annually and cut operational costs by ~50% by filtering out low-value calls before they reach the CRM.

πŸ† Outcome: Won the hackathon for innovative approach, rapid execution, and demonstrable business value in just 4 hours!


πŸ† Hackathons

πŸ€– Google Agent Development Kit (ADK) Hackathon - June 2025

Oprina - Conversational AI Avatar Assistant
Built a revolutionary voice-powered AI assistant combining conversational intelligence with interactive avatar technology. The platform transforms email management, calendar scheduling, and productivity tasks into natural voice conversations.

πŸ—οΈ Architecture: 5-component microservices (Frontend, Backend, AI Agents, Database, External APIs) with specialized Email and Calendar agents powered by Google's Agent Development Kit and deployed on Vertex AI.

πŸ› οΈ Tech Stack: Google ADK, Vertex AI, React 18 + TypeScript, FastAPI + Python, Supabase, Google Cloud Run, HeyGen Streaming Avatars, Gmail & Calendar APIs, Speech-to-Text & Text-to-Speech

🎯 Outcome: Didn't win, but gained incredible experience building and deploying a complete platform in under a month!

πŸ”— Links: Live Demo | GitHub Repository


πŸš€ Featured Projects

πŸŽ™οΈ AUXIS AI: Voice-First Chrome Extension

Built a Chrome extension eliminating copy-paste workflows between ChatGPT and Gmail/LinkedIn β€” compose messages directly in-place using voice.

Key Features:

  • Real-time WebSocket streaming transcription
  • Scenario-based AI text generation (OpenAI/Gemini)
  • Context extraction for reply threads
  • Voice dashboard with conversational AI agent

Tech: Chrome Extension (Manifest V3), FastAPI, Google Cloud STT/TTS, Vertex AI, OpenAI, Gemini, Supabase, Cloud Run

Role: Solo Developer

πŸ”— Links: Live Demo

🧠 Agentic Corrective RAG System for Research Paper Queries

A production-ready RAG system that synthesizes academic content using multi-retriever setups, LLM-based document evaluation, and smart query rewriting with web search integration.
Tech: LangChain, LlamaIndex, Faiss, OpenAI, PostgreSQL

πŸ”€ English-to-Telugu Translation using Mistral 7B

Fine-tuned a Mistral 7B Instruct model with QLoRA for high-quality English–Telugu translation using a 140k sentence-pair dataset.
Tech: PyTorch, Transformers, Hugging Face, QLoRA

🌍 pygeoboundaries

A Python package developed at William & Mary for validating and reviewing geoBoundaries shapefiles. Published on PyPI and integrated into open-source workflows.
Tech: Python, Prefect, Web Scraping, Dataverse API


🌟 Contributions & Community Engagement

  • Published open-source tools like pygeoboundaries
  • Mentored undergraduates in AI and ML applications at Texas A&M University

πŸ“« Let’s Connect

I’m always open to collaboration in Generative AI, agentic systems, or cutting-edge RAG applications.

πŸ“« How to reach me:

Pinned Loading

  1. IFT-for-English-Telugu-Translation IFT-for-English-Telugu-Translation Public

    Jupyter Notebook

  2. Agentic-CRAG-System Agentic-CRAG-System Public

    A Research Paper Answer Bot with Agentic Corrective RAG, vector database, and multi-retriever setup

    Jupyter Notebook 1

  3. FT_LLMs FT_LLMs Public

    Jupyter Notebook

  4. Oprina_AVAFG Oprina_AVAFG Public

    Oprina is an AI-powered conversational agent with real-time avatar streaming, featuring multimodal Gemini 2.0 Flash integration, Gmail/Calendar automation, voice controls, and enterprise-grade dep…

    Python 8 3

  5. wmgeolab/geoBoundaries wmgeolab/geoBoundaries Public

    geoBoundaries : A Political Administrative Boundaries Dataset (www.geoboundaries.org)

    411 69

  6. DanRunfola/dataSecChallenge DanRunfola/dataSecChallenge Public