Specialized in developing cutting-edge ML/DL architectures with 5+ years of experience. From computer vision to NLP, I architect solutions that solve real-world problems at scale.
Hi, I'm Carlo Calledda. I started with a thesis on neural networks for seismic design and academic research in data science and machine learning engineering, which I turned into specialized experience developing end‑to‑end MLOps solutions and computer vision systems.
Today I'm a Data Scientist specialized in MLOps implementations and microservice systems: I design automated ML pipelines, build REST APIs with FastAPI, and deliver containerized deployments. I have proven experience delivering end‑to‑end solutions, from dataset automation tools to production‑ready services.
I have a solid foundation in structural engineering and growing expertise in NLP and large language model applications, with a strong passion for Rust. I aspire to lead projects that combine my MLOps background with emerging ML/NLP/LLM technologies.
High-performance Rust implementation of GraphRAG that builds knowledge graphs from documents and enables natural language queries. ⭐ 230 stars
Lightweight AI agent orchestrator built in Rust. High-performance Actix-web backend for agent lifecycle management and a real-time WASM dashboard.
Reinforcement Learning trainer for LLM function calling using GRPO. AST-based reward functions, QLoRA quantization optimized for consumer GPUs.
Blazingly fast chat interface built with Rust/WASM and Leptos, featuring local AI model execution via WebLLM. Privacy-first, no backend. ⭐ 12 stars
WASM-powered Typst Studio with real-time compilation to SVG/PDF, IEEE template support, and dynamic bibliography management. ⭐ 41 stars
MCP server that transforms Rust documentation into a queryable knowledge graph, enabling semantic search and AI-assisted discovery.
MCP server for parsing and querying Rust documentation. Extracts details about functions, structs, and enums via a programmable interface.
MCP server that analyzes Python projects, extracts library signatures, and surfaces project information to help AI assistants understand codebases.
High-performance object detection service built with Rust and Actix-Web, focused on robustness and production-ready APIs.
Web-based image editor built with Rust & WebAssembly. Apply filters, adjust colors, and process images in real-time with near-native performance.
Fast tool to parse Rust projects into a knowledge graph, run queries over code structure, and generate DOT/SVG visualizations.
WebAssembly-based chat application with local LLM execution in the browser. Privacy-focused with client-side inference.
Vector presentation format with synced audio. A .vecslide ZIP contains SVGs and Opus audio via YAML manifest, compiling to standalone zero-dependency HTML. ⭐ 3 stars
Chrome extension (MV3) and multi-agent SRE/DevOps copilot. Offline YAML validation and auto-repair for Kubernetes, Docker Compose, GitLab CI, GitHub Actions, Prometheus.
Agentic SRE Assistant: local-first AI for log analysis, doc RAG, and YAML validation. Built with PydanticAI, DuckDB, and Apache Iceberg.
Pure-Rust data visualization library using Leptos and WebAssembly. Automatic Level of Detail management, rendering massive datasets via LTTB and M4 downsampling. ⭐ 5 stars
Multi-LoRA fine-tuning of IBM Granite 4.0: 3 specialized adapters (graphrag/synthesizer/agent) on a single 3B base with zero-cost adapter swapping.
Automates creation of production-grade synthetic datasets for function-calling LLMs by ingesting OpenAPI schemas and utilizing local inference via Ollama.
100% offline, privacy-focused GraphRAG system built in Rust for local document intelligence on consumer hardware.
Production-grade mini-FaaS/Lambda system with WASI 0.2 and Component Model. ~450 LOC delivering hot-reload, Prometheus metrics, Redis/PostgreSQL integration.
Privacy-first, client-side data visualization powered by Rust + WebAssembly. 11 chart types, drag-and-drop layout, auto-save, undo/redo — all in-browser. ⭐ 8 stars
Neural-network inversion framework to optimize structural parameters for seismic resilience. Combines simulated ground-motion data, sensitivity analysis and gradient-based search to minimize expected seismic demand.
Built an astronomy image dataset with Aladin, applied targeted data augmentation, and bootstrapped a detector from an image classifier. Optimized for low-power CubeSat hardware via pruning and quantization.
Self-Organizing Map approach on multi-spectral remote-sensing imagery to detect marine plastic aggregations, including atmospheric correction, texture features and spatial post-processing.
Job-ready training in spreadsheets, SQL, R, and data visualization.
Python scripting, Git/GitHub, troubleshooting, and automation of IT tasks.
Data modeling, ETL, SQL, and dashboarding for business-ready insights.
Supervised ML, classification, regression and neural networks.
RNNs, attention, transformers and sequence-to-sequence models.
End-to-end ML systems: data pipelines, deployment, monitoring, and CI/CD.