Aryan Vaidh
About Candidate
I am an AI ML Engineer with 2 years of production experience specializing in Generative AI, Retrieval-Augmented Generation (RAG) pipelines, and autonomous multi-agent workflows. I have a proven track record of bridging machine learning research with scalable SaaS architectures. My core expertise lies in architecting Human-in-the-Loop (HITL) systems using LangGraph, fine-tuning LLMs for production, and developing ultra-low latency real-time AI voice agents. I am highly proficient in Python, FastAPI, vector databases, and deploying robust backend microservices on AWS and GCP.
Location
Education
Work & Experience
Building production-grade AI systems and shipping them as real SaaS products.
● Proprietary AI Humanizer - End-to-End SaaS Product
- Fine-tuned Llama-3 (8B) using Unsloth (PEFT/QLoRA) with 4-bit quantization on a custom 40,000-row dataset to bypass commercial AI detectors.
- Deployed on AWS GPU via vLLM with async SSE streaming for real-time inference. Architected a FastAPI billing proxy with token-based subscriptions and payment gateway integration.
- Built the entire product solo - UI/UX, Node.js frontend, and backend from zero to paying customers.
● Real-Time AI Voice Calling Agent & Campaign Dashboard
🚀 Live Dashboard: https://ai-voice-agent-dashboard-ashy.vercel.app
- Architected an end-to-end autonomous voice AI system capable of conducting human-like outbound phone calls for automated lead generation and customer outreach.
- Built the real-time audio pipeline using Python, FastAPI, and the Pipecat framework, streaming bidirectional low-latency audio over WebSockets via Vobiz telephony.
- Integrated Google Gemini for conversational intelligence, alongside dynamic STT/TTS models, enabling the agent to handle interruptions and context-aware responses seamlessly.
- Developed a full-stack Next.js dashboard backed by Supabase (PostgreSQL) to manage campaign lead data, automate continuous outbound dialing loops, and provide real-time call monitoring with full conversation transcripts and cloud-hosted audio recordings.
● Autonomous Content Strategy Agent
- Built a Human-in-the-Loop (HITL) multi-agent pipeline using LangGraph + Google Gemini, orchestrating four specialized agents (Analyst, Strategist, Writer, Publisher) triggered by a single Slack keyword - delivering fully drafted content directly to Wix CMS via REST API.
- Integrated Serper.dev for real-time competitor scraping, enforced structured JSON outputs via Pydantic, and maintained SQLite graph state persistence across restarts - reducing manual research and publishing time by 80%.