Pranav Budhwant

About Pranav Budhwant

I build AI systems end-to-end — from model training and fine-tuning to production deployment and customer-facing iteration. My work spans ML Engineering, Agentic AI, Computer Vision, Large Language Models (LLMs), Retrieval-Augmented Generation (RAG). I\'m most energised at the 0→1 stage: taking a problem through system design, rapid prototyping, and into production — often working directly with customers along the way.Current focus areas- Agentic AI systems — multi-agent orchestration, tool-calling, LLM pipelines (LangChain, LangSmith, OpenAI, Ollama)- Multimodal AI — vision-language models, video understanding, embedding-based retrieval, RAG & Vector Infrastructure — Qdrant, dense/sparse retrieval, multi-scale embedding pipelines- ML Systems — model training, fine-tuning (YOLOv9, LLaVA, CLIP variants), inference optimization, WandB, SageMaker, AWSI have production experience across the full ML stack: data pipelines, model iteration, distributed inference, vector DBs, and system design for real-time and batch workloads.On the side, I\'m actively exploring Video AI Agents and Voice Agents — building at the intersection of multimodal perception and autonomous decision-making.Open to Forward Deployed, Software, ML Engineering, Applied AI, and Agentic Systems. Let\'s connect.

Experience

1. Software Engineer - Machine Learning

   Dragonfruit Ai

   Jan 2025 — Present · Los Angeles, CA, US

   Building and deploying production ML systems across computer vision, LLMs, and agentic AI.Text-Based Video Search & Alerting- Embedding-based video retrieval system supporting semantic search over large-scale camera fleets- Real-time free-text alerting pipeline using open-vocabulary object detection for zero-shot event classification- Distributed query routing infrastructure using Redis, FRP tunneling, and REST/RPC, with load-aware node selection- Stack: Python, PyTorch, Qdrant, Elasticsearch, Flask/FastAPI, Redis, Docker, AWSMulti-Agent Go-To-Market System- CRM-integrated multi-agent pipeline automating lead enrichment, product recommendation, and personalized outreach at scale- Built LangChain-based agents with tool-calling (web search), HubSpot MCP integration, and Qdrant-based Retrieval Augmented Generation (RAG) for per-contact context retrieval and outreach- Optimized inference cost via batching, prompt caching, and on-prem inference offloading across a distributed Ollama cluster- Stack: LangChain, LangSmith, OpenAI, Ollama, Qdrant, HubSpot MCP, Python

Education

• Pune Institute of Computer Technology

  Bachelor of Engineering, Computer Engineering

• University of Southern California

  Master of Science - MS, Computer Science