Detailed builds across robotics, AI/ML, and backend systems with architecture, product decisions, and implementation notes.
Deeper public-safe breakdowns are available in the case studies section.
Multi-tenant AI sales platform with a production backend, dashboard, and mobile app — automating customer conversations and connecting them to order, invoice, and payment workflows.
Tenancy
Business-scoped access
Channels
Social inbox automation
RAG-based document Q&A app — upload PDFs/text, ask questions, get answers with chunk-level citations. Runs in mock mode without paid LLM keys.
Pattern
Retrieve · Cite · Generate
Run mode
Mock or OpenAI-backed
Webcam / demo-feed object-detection dashboard with FastAPI WebSocket streaming, YOLO-ready detector abstraction, canvas bounding-box overlays, and live throughput / latency / confidence stats.
Stream
WebSocket + backpressure loop
Detector
Mock or YOLOv8 adapter
Event ingestion + live inspector: receive provider callbacks, browse payloads in real time, replay captured events, and forward matches with bounded retries.
Pattern
Ingest · Inspect · Replay · Forward
Reliability
Verified ingest + retries
Interactive 2D path-planning sandbox — paint obstacles, drop start/goal, and compare A*, Dijkstra, and an RRT-style planner with visited-cell visualization and live stats.
Planners
A* · Dijkstra · RRT-style
Stats
Cost · Visited · Runtime · Success
Adaptive music recommendation system that learns listener preferences over time and generates explainable playlists using embeddings and contextual signals.
Modeling
taste embeddings + memory
Output
explainable playlist ranking
Autonomous retail robot for inventory tracking, customer assistance, and ERP synchronization using ROS, YOLOv5, and DeepSORT.
Detection
YOLOv5 + DeepSORT
Training
1000 epochs
ROS-based quadruped guidance system using YOLO perception and audio feedback to support visually impaired users in navigation tasks.
Platform
Unitree Go1 + Gazebo
Perception
real-time YOLO pipeline
ROS and Gazebo disaster-response simulation project focused on SLAM mapping, autonomous navigation, and hazard-aware exploration.
Environment
unknown-map exploration
Pipeline
SLAM + autonomous routing
Applied AI + IoT project using neural networks to predict photovoltaic power efficiency from real-time operational telemetry.
Training
1000 epochs
Evaluation
MSE, MAE, MAPE, R²
Evolutionary optimization project implementing selection, crossover, mutation, and fitness-driven search.
Algorithm
selection + crossover + mutation
Use case
complex search-space optimization
Portfolio of data mining and supervised learning experiments covering preprocessing, feature engineering, and predictive modeling.
Scope
multi-dataset experiments
Focus
end-to-end ML workflow