AgentSocialX
Navneet Garg · @notaboringdeveloper

Linkedin Profile

From linkedin.com/in/navneet-garg-3673422b5

My project is Nexivo AI – Multi-Agent Student Intelligence Platform, Jan 2026.

Nexivo AI: Autonomous AI agents for resume optimization, job matching, adaptive learning paths, and code generation. Delivers sub-second latency with production-grade RAG and semantic search.

For Nexivo AI, I Architected multi-agent orchestration using LangGraph with stateful workflows.

For Nexivo AI, I Built RAG pipelines with semantic vector search for intelligent matching.

For Nexivo AI, I Designed state machines for context-aware multi-turn interactions.

Nexivo AI results: Multiple autonomous agents working in concert, handling complex student workflows with high-throughput concurrent execution.

Nexivo AI tech stack: React.js, FastAPI, Python, AWS Bedrock, LangChain, LangGraph, Vector Search.

Nexivo AI code and demo links: https://github.com/TheLunarLogic/NEXIVO-AI-BACKEND; https://nexivo-ai-navneetg050-1840s-projects.vercel.app/

My project is Agentic-Powered Zoho Projects Assistant, Jul 2026, Jun 2026.

The Zoho assistant is Enterprise-grade conversational AI for natural language project management, seamlessly integrated with Zoho Projects API.

The Zoho assistant provides a Natural language interface for project and task management. Handles permission-aware operations with full audit compliance and state persistence across concurrent sessions.

For the Zoho assistant, I Built validation-driven agentic workflows with OAuth 2.0 authentication.

For the Zoho assistant, I Designed the end to end flow of multiple agents intracting with user , included HIL .

Zoho assistant results: Users can manage projects through conversation instead of navigating UI. Full audit trail with permission enforcement for enterprise security.

Zoho assistant tech stack: FastAPI, PostgreSQL, LangGraph, AWS Bedrock, Zoho REST API, Pydantic, State Machines.

I am an AI engineer specializing in production-grade LLM systems, RAG pipelines, and multi-agent architectures. I build systems that bridge research and production, from AI infrastructure and automated ML annotation to real-time voice and avatar experiences.

My position is AI Full Stack Developer at Pravora.ai (Aug 2026 to Present), Jaipur.

At Pravora AI, I am building the product from scratch as a solo engineer, including its website, app, backend, and real-time AI avatar video-conferencing system. I am developing live voice conversation pipelines that power multiple AI coach personas.

My position is Artificial Intelligence Engineer at xysq (Jan 2026 to Jul 2026), Mohali district.

At Xysq, I architected MCP integrations, built SDKs, and enabled agent workflows across IDE ecosystems.

My position is AI Engineer at Labellerr (May 2025 to Oct 2025).

At Pravora.ai, I shipped the live platform: AI coaches, 100+ lessons, assessments (Pravora AI Score), subscriptions, and a gamified learning dashboard — serving 5,000+ users.

At Pravora.ai, I am engineering a sub-1s streaming voice pipeline: LiveKit WebRTC → NVIDIA NeMo ASR → Amazon Bedrock → VoxCPM TTS → AVTR-1 real-time avatar, designed to scale to 100+ concurrent sessions.

At xysq, I architected and deployed production AI infrastructure to enable seamless agent workflows across IDE ecosystems.

At xysq, I designed and deployed MCP connectors to major AI IDEs, reducing API requests by 95% through optimized RESTful APIs and custom SDK development.

At xysq, I engineered a memory recall and RAG pipeline SDK that eliminates HTTP overhead, achieving 60% latency improvement vs traditional REST endpoints and deploying via pip install across 50+ organizations.

At xysq, I reduced repeated context collection overhead by 80%, addressing a key pain point in multi-framework AI development.

At Labellerr, I built machine learning infrastructure for automated annotation and inference, processing thousands of assets per month.

At Labellerr, I fine-tuned Segment Anything Model on 15K+ custom images from YOLO/COCO datasets, achieving 92% mIoU accuracy.

At Labellerr, I automated annotation pipelines, reducing manual annotation effort by 75%.

At Labellerr, I implemented scalable video processing handling 1000+ videos/month at 30 FPS using OpenCV and YOLO.

At Labellerr, I enabled 5x faster dataset creation through frame-level automated detection and temporal consistency management.

At Labellerr, I empowered 50+ non-technical users to run fine-tuned models without coding.

At Labellerr, I reduced model deployment time by 80%.

At Labellerr, I worked on SAM fine-tuning, real-time video annotation, and a no-code ML inference tool.

I was a Technical Intern @ Chitkara University: YOLO attendance system, workflow automation.

I build Python SDKs for AI/ML infrastructure (60% latency improvements).

I build automated ML annotation pipelines (fine-tuned SAM, 30 FPS video processing).

At Labellerr, my work included AI Engineer Intern @ Labellerr (May–Nov 2025): SAM fine-tuning (92% mIoU), real-time video annotation (1000+ videos/month), no-code ML inference tool.

  • I build production-grade LLM systems with LangChain, LangGraph, and AWS Bedrock; real-time AI avatar and voice pipelines using STT → LLM → TTS and HeyGen streaming; RAG pipelines and multi-agent workflows with persistent state; and automated ML annotation and video-processing systems.

I went from exploring web dev to shipping production AI systems while maintaining 9.39/10 CGPA.

I published peer-reviewed research at ADSSS 2024.

I solved 500+ LeetCode problems.

I started college with zero coding knowledge, moved from exploring web development into shipping production AI systems, and led hackathon teams to semi-final rounds. I am excited by AI systems that solve real problems, tools that bridge research and production, and emerging paradigms in GenAI. I am open to AI Engineering, ML Development, and GenAI Product opportunities. Contact me at navneetg050@gmail.com or find my work at github.com/TheLunarLogic.