I am an AI Engineer and Software Developer focused on building production-grade Generative AI systems, real-time voice-to-voice pipelines, and scalable full-stack platforms. My current work is deeply centered around low-latency audio intelligence, multi-agent architectures, and applied LLM systems operating under real-world constraints.

I have hands-on experience designing end-to-end V2V (voice-to-voice) AI pipelines, telephony-integrated conversational agents, and multi-agent automation frameworks used in enterprise and consumer-facing environments. Alongside AI systems, I actively build high-performance backend services and modern web applications with an emphasis on reliability, observability, and scalability.

I regularly document my engineering learnings, experiments, and architectural decisions publicly on X, particularly around real-time audio AI, LLM orchestration, and system-level optimizations.

• Real-time Voice AI systems (STT → LLM → TTS, V2V pipelines)
• WebRTC, VoIP, and telephony-based conversational agents
• Multi-agent LLM orchestration using LangGraph and LangChain
• Low-latency, cost-aware model routing and inference optimization
• Scalable backend systems with async-first architectures
• Applied RAG systems with vector databases

Aug 2025 – Present

• Architected and deployed an end-to-end Voice-to-Voice AI pipeline from scratch using OpenAI GPT-4o-mini Realtime Preview, Gemini native thinking-audio models, Whisper ASR, and custom STT–LLM–TTS micro-pipelines.
• Achieved sub-220ms round-trip latency via chunked audio streaming, adaptive buffering, server-side VAD tuning (60–80ms), and preemptive interruption handling.
• Integrated DeepFilterNet for real-time noise suppression, improving speech clarity by approximately 32 percent.
• Designed a real-time communication stack using LiveKit (WebRTC), WebSocket-based audio streaming, and VoIP protocol experimentation (μ-law, Opus, RTP jitter buffers).
• Built bidirectional telephony integrations with Twilio and Exotel, including custom G711 μ-law converters, FFmpeg preprocessing, and async worker pipelines, resulting in 98 percent plus call stability.
• Conducted R&D across OpenAI, Gemini, Anthropic, Deepgram, and Groq voice models with token-aware routing, latency-based model switching, and compression-optimized payloads, reducing per-call cost and improving inference throughput by ~35 percent.
• Designed PostgreSQL-backed async architectures with multi-stage NLP extraction and multi-channel support (Telephony, WhatsApp, Telegram).

Jun 2024 – Aug 2024

• Developed optimized PostgreSQL query pipelines for large-scale data extraction and reporting.
• Improved reporting efficiency by ~30 percent for clients managing 100,000 plus records.

• Designed an enterprise-grade AI automation platform using LLMs, FastAPI, Docker, Pinecone, and LangChain orchestration.
• Deployed autonomous micro-agents for email automation, voice-based FAQ and data collection, WhatsApp outreach, and analytics dashboards.
• Implemented semantic search, real-time telephonic interactions, and scalable messaging infrastructure.
• Achieved approximately 50 percent operational efficiency gains and 40 percent higher stakeholder engagement.

• Architected a production-grade multi-agent system using LangChain, LangGraph, and FastAPI.
• Orchestrated five specialized agents with asynchronous coordination and persistent state management.
• Implemented enterprise-grade practices including structured logging, exception handling, CI/CD pipelines, UV package management, and Husky git hooks.
• Integrated multiple LLM providers (OpenAI, Groq) with concurrent asyncio workflows, achieving ~95 percent uptime.

• Built an AI-driven SaaS platform that converts text documents into summarized, voice-over-enabled videos with multilingual support.
• Integrated analytics dashboards, interactive quizzes, and engagement tracking.
• Added inclusive features such as custom knowledge-based assistants, AR/VR components, sign-language video generation, and voice-controlled navigation.

Programming Languages

• Python, C++, JavaScript, TypeScript, SQL

AI and Generative Systems

• LangChain, LangGraph, LangSmith, RAG pipelines, LLM embeddings, Pinecone
• Speech and audio systems, STT/TTS pipelines, LiveKit

Backend and Infrastructure

• Node.js, Express.js, FastAPI, Kafka, Redis
• Docker, PostgreSQL, MongoDB, Firebase, SQLite

Frontend and Web

• React, Next.js, Redux, Tailwind CSS

Tooling and Platforms

• Git, GitHub Actions, CI/CD pipelines, Selenium, BeautifulSoup, GCP

• Top 22 National Finalist and Social Buzz Winner — Bajaj Finserv HackRx 5.0
• Winner — Aditya Birla Group Synaptix Hackathon 2025 (DTU)
• Selected — Microsoft GitHub Field Day, Gurgaon
• First Runner-Up — FusionFest Hackathon, Chitkara University
• Finalist and Social Buzz Winner — Vihaan 7.0, DTU

• Portfolio: https://www.yadavhappy.in
• LinkedIn: https://www.linkedin.com/in/happy-yadav-16b2a4287
• Email: [email protected]

I am interested in roles and collaborations involving Generative AI, Voice AI, LLM Systems, and scalable backend engineering. Open to research-driven engineering problems and early-stage product work.