Shikhar Srivastava — Portfolio 2.0

AI/ML Engineer specializing in production-grade agentic AI, real-time voice pipelines, computer vision, time-series forecasting, and high-performance async backends.

Live Site: shikhar-portfolio-phi.vercel.app

---

Shipped Production AI Systems (Zypp Electric)

During my tenure at Zypp Electric (India's largest EV-first last-mile delivery platform), I built and shipped 8 core production AI systems serving 25,000+ active delivery riders daily:

1. Zypp Saathi (Voice AI): A real-time multilingual voice assistant built using Pipecat, OpenAI Realtime API, and LiveKit WebRTC. Deflects 40%+ of daily support calls by handling earnings, navigation, and onboarding queries over voice in Hindi and Indic languages.
2. Zia Chatbot (Phase 3.0): Conversational agent runtimes migrated to LangGraph state machines with Qdrant Vector DB retrieval, resolving 50%+ of daily worker support tickets automatically.
3. AI Vehicle Validation: A GPT-4o Vision damage validation pipeline analyzing 4 vehicle sides in parallel for ride-ends, saving 100+ manual operations hours/day.
4. Zybil / Hustle Score (Credit Engine): A proprietary 300–900 point credit scoring engine tracking EV IoT telemetry, wallet health, and merchant performance data to automate loan eligibility with a 0.0% default rate.
5. Spare Parts Forecaster: Consumption demand forecasting across 15+ cities using Amazon Chronos-2 with TSB/SBA statistical fallbacks.
6. Fraud Detection & Blacklisting: Verification pipeline using MediaPipe and AWS Rekognition face collections blocking 99.8% of duplicate driver registrations.
7. Agentic Lead Generator: AI-agentic chatbot utilizing FastMCP tool servers for CRM writes, boosting qualification rates by 28%.
8. EV Component OCR: High-accuracy (99.4%) UID and 15-digit IMEI extraction engine optimized via Nvidia Nemotron-Parse-1.1.

---

Tech Stack & Architecture

• Backend & Core: Python (Async), FastAPI, SQLAlchemy, MySQL, Docker, NGINX
• AI/ML & Audio: Pipecat, LiveKit (WebRTC), OpenAI (Realtime, GPT-4o Vision), Amazon Chronos-2, MediaPipe, AWS Rekognition, Nvidia Nemotron-Parse
• Vector Search & Agents: Qdrant Vector DB, LangGraph, FastMCP (Model Context Protocol)
• Caching & Scheduling: Redis, APScheduler / ARQ
• Frontend: Next.js 16 (Turbopack), TypeScript, Framer Motion, Vanilla CSS

---

Getting Started

Prerequisites

• Node.js (v18+)
• Python (v3.10+)

Setup Instructions

1. Clone the repository:

   git clone https://github.com/Shikhaar/Portfolio2.0.git
   cd Portfolio2.0

2. Install dependencies:

   npm install

3. Configure Environment Variables: Create a .env.local file in the root directory:

   OPENROUTER_API_KEY=your_key_here
   MODEL_NAME=google/gemini-2.5-flash

4. Generate Search Embeddings (For Chat Panel Context):

   npm run generate-embeddings

5. Run the development server:

   npm run dev

---

License

This project is licensed under the MIT License.