AI Systems Engineer | Robotics and AI Systems

Viplav Dodeja

Building intelligent systems across AI infrastructure, multimodal workflows, and real-world applications.

Explore Projects Contact

Embodied AI

Persistent memory

Edge-cloud robotics

Agent automation

Builder Philosophy

Systems first. Product always.

I build systems where AI becomes operational: edge robotics, persistent memory, multimodal infrastructure, and the infrastructure needed to turn prototypes into durable products. As both an engineer and startup founder, I enjoy fast iteration and building systems end-to-end. My work ranges from self-managed GPU infrastructure and backend orchestration to multimodal robotics platforms designed for real-world interaction.

Project Index

Start with the signal. Go deeper when it matters.

A fast scan layer before the deeper system case studies.

VIP Twin

Persistent memory AI system

Local-first AI memory with durable context, structured/vector retrieval, and route-aware orchestration.

FastAPIMySQLQdrant

Explore project->

NOVA

AI robotics + edge intelligence

Robotics intelligence that keeps perception local, routes heavier reasoning to cloud services, and turns inference into motion.

Raspberry PiArduinoOpenAI APIs

Explore project->

PhantomPay

Adaptive AI workflow prototype

AI-driven invoice workflows using specialized agents, adaptive reminders, and feedback loops.

TypeScriptNode.jsPostgreSQL

Explore project->

Smart Pantry

Inventory forecasting + planning

Applied AI system for shared residential inventory visibility, grocery forecasting, and adaptive planning.

Recommendation systemsForecasting workflowsInventory data

Explore project->

Nemesis Industries

Product design engineering

Off-road armor and aftermarket accessory development for the Toyota Tundra platform.

SOLIDWORKSCreaform GO!SCAN SparkVX Scan

Explore project->

Flagship Project

VIP Twin

Memory-native AI for persistent personal context.

Profiles, conversations, documents, and structured memory combine into a durable intelligence layer.

Persistent MemoryDocument IntelligenceProfile PersonalizationLocal ModelsQdrantFastAPI

Memory Core

VIP Twin

A personal AI layer that remembers profiles, conversations, facts, and documents across sessions.

MySQL canonical memory

Qdrant retrieval pointers

Ollama local inference

Profile

Chat

Documents

Memory

Retrieval

Product Architecture

A memory system, not a chat wrapper.

Durable profiles, documents, memory facts, and selective retrieval keep context useful over time.

VIP Twin memory and retrieval architecture diagram

Memory Source of Truth

MySQL stores canonical profiles, conversations, documents, and durable facts.

Vector Retrieval Layer

Qdrant holds embeddings and pointers for fast contextual recall.

AI Orchestration Layer

FastAPI coordinates streaming, routing, grounding, and retrieval policy.

Local-First Inference

Ollama and self-hosted services keep inference private and portable.

Live App

Check out VIP Twin yourself.

Explore the app directly and see the persistent-memory product outside the portfolio.

Open viptwin.org

Flagship Project

NOVA

A Hybrid Edge-Server AI Architecture for Real-Time Robotics

Computer vision, voice interaction, server reasoning, and actuation in one embodied robotics loop.

3rd Place, 2026 Fremont Tech Week Capstone Expo

Remote Linux server + Dockerized services

Edge AIRoboticsComputer VisionMultimodal AIRaspberry PiEmbedded SystemsRedis MessagingDockerized Services

Watch the demo

Vision online

Pi edge active

Arduino linked

Multimodal AI Pipeline

From human signal to physical response.

NOVA multimodal edge-server robotics pipeline diagram

Live Demo

Hardware, interaction, and control in one loop.

The fastest read on NOVA: a real robot responding through the edge-cloud pipeline.

Real robot demo

Edge-cloud workflow

Fast web playback

System Overview

Local reflexes. Cloud-scale reasoning. One robotic loop.

Edge, actuation, and server layers keep the robot responsive while Redis-backed services handle heavier inference.

low-latency control

Raspberry Pi Edge Layer

Runs live sensing, wake-word input, YOLOv11 perception, and local routing.

motor + servo execution

Arduino Actuation Layer

Executes serial commands for motor direction, PWM speed, servo angle, and movement timing.

cloud-assisted intelligence

Server Reasoning Layer

Offloads contextual interpretation, extended responses, and future memory workflows.

voice + vision + motion

Multimodal AI Pipeline

Links camera input, speech, AI interpretation, TTS, and physical response.

Operational Use Cases

A modular embodied AI framework, not a single-purpose robot.

NOVA's edge-server architecture can adapt to multiple operational environments where perception, interaction, and physical response need to work together.

Modular Robotics Framework

reusable embodied AI architecture

Edge-server orchestration, distributed intelligence, and adaptable multimodal workflows.

Warehouse Intelligence

mobile operational perception

Inventory observation, asset tracking, environmental monitoring, and anomaly detection through edge vision.

Assistive Robotics

human-centered interaction

Voice interaction, reminders, follow-me behavior, and contextual awareness for interactive assistance.

Campus Interaction

sensing + operational support

Attendance workflows, engagement monitoring, campus assistance, and behavioral analytics.

Core Features

Embodied AI capabilities, compressed.

NOVA sees, listens, routes, speaks, tracks, and moves through coordinated edge/cloud subsystems.

Real-Time Object Detection

YOLOv11 processes live webcam frames for object and person awareness.

Hybrid AI Reasoning

Local perception handles latency; server reasoning handles context.

Voice-Controlled Movement

Wake-word commands map intent into movement primitives.

Follow-Me Tracking

Frame-center tracking drives alignment and movement correction.

Servo Camera Tracking

Servo-mounted vision keeps targets centered during perception tasks.

Scene Narration

Visual detections become spoken environmental descriptions.

Future Vision

Toward distributed embodied intelligence.

Next steps: ROS2 control, SLAM navigation, persistent memory, and multi-agent coordination.

ROS2 integration

Modular control and cleaner subsystem communication.

SLAM navigation

Mapping and localization beyond calibrated motion.

Multi-agent robotics

Shared perception and task context across robots.

Persistent AI memory

Session continuity for embodied interaction.

Distributed intelligence

Edge devices and cloud reasoning as one system.

Projects

Additional systems work.

PhantomPay Smart Pantry Nemesis Industries

03 / Adaptive AI workflow prototype

PhantomPay

Adaptive AI workflow prototype

AI-driven invoice workflows using specialized agents, adaptive reminders, and feedback loops.

Discuss a similar system

47-second product demo

Agent workflow automation

Invoice follow-up loop

TypeScriptNode.jsPostgreSQLDrizzle ORMGPT-4o-mini

Workflow architecture

RiskScoring, ReminderStrategy, Coordinator, and EmailComposer agents

Bandit-style strategy selection shaped by payment outcomes

Contextual prompt injection for customer-specific communication

Explainable risk scoring and modular backend orchestration

Engineering challenge

Coordinate risk scoring, reminder strategy, tone calibration, and email generation in one adaptive workflow.

Impact

A rapid prototype for overdue-payment automation with strategy selection and payment-outcome feedback.

04 / Inventory forecasting + planning

Smart Pantry

Inventory forecasting + planning

Applied AI system for shared residential inventory visibility, grocery forecasting, and adaptive planning.

Discuss a similar system

Inventory visibility

Forecasting workflow

Adaptive grocery planning

Recommendation systemsForecasting workflowsInventory dataDashboardsFull-stack app

Planning workflow

Resident inventory tracking and visibility dashboards

Consumption analysis for future grocery need forecasting

Adaptive recommendation workflows shaped by preferences

Automated grocery list generation for shared food planning

Engineering challenge

Turn household inventory, consumption behavior, and resident preferences into usable planning workflows.

Impact

A prototype for reducing waste, improving visibility, and automating grocery list generation.

05 / Product design engineering

Nemesis Industries

Product design engineering

Off-road armor and aftermarket accessory development for the Toyota Tundra platform.

Discuss a similar system

CoreNemesis Industries

3D scanning

CAD modeling

Fitment validation

Fabrication feedback

SOLIDWORKSCreaform GO!SCAN SparkVX ScanVX ModelProduct prototyping

Engineering scope

Industrial 3D scanning workflows for accurate vehicle geometry

SOLIDWORKS modeling from scan data and fitment constraints

Prototype refinement with durability, weight, and structure in mind

Collaboration across design, fabrication, and validation cycles

Engineering challenge

Turn vehicle scan data, fitment constraints, and fabrication feedback into manufacturable designs.

Impact

Improved prototype precision, vehicle fitment, and manufacturing readiness across real hardware iterations.

Early-stage AI products, built as systems.

Memory that persists, robots that perceive, and workflows that turn reasoning into action.

Founder building VIP Twin

Local-first AI memory platform built through the Innovate Bay IB-1 cohort.

Founding engineer at Sparkk

Pre-launch AI education MVP using FastAPI, LLM APIs, and modular services.

Robotics coach and instructor

Hands-on robotics instruction across engineering, code, and competition strategy.

Product engineering foundation

CAD, prototyping, deployment, cloud, and embedded systems across real builds.

My Skills

Systems I build with.

AI Systems

OpenAI APIsLLM OrchestrationRAGMultimodal AIEmbodied AISpeech-to-TextVector Search

Robotics

Raspberry PiArduinoOpenCVYOLOv11Servo ControlUART CommunicationEdge Robotics

Infrastructure

DockerRedisLinuxSystemdCaddyEdge-Server Architectures

Frontend

Next.jsReactTypeScriptTailwindFramer Motion

Backend

FastAPINode.jsMySQLPostgreSQLQdrantWebSocketsAsync Systems

Cloud & DevOps

GCPIBM CloudSSHRemote DeploymentGPU InfrastructureCloud Orchestration

CAD & Product Design

SOLIDWORKSIndustrial 3D ScanningCreaform GO!SCAN SparkVX Scan / VX ModelPrototype Fitment Validation

Contact

Let's build something intelligent.

Robotics, AI memory infrastructure, agent workflows, and prototypes with a path to deployment.

Open a Conversation LinkedIn GitHub

viplav.dodeja@gmail.com +1 (510) 456-8101