Clelia Astra Bertelli

Articles on coding agents, document workflows and AI infrastructure.

Author of technical articles for LlamaIndex and independently on:

• Coding agent sandboxing
• Skills vs MCP tools
• Observability for AI systems ([1], [2])
• AI agents and the tooling around them ([1], [2])
• Filesystems, search and RAG ([1], [2])
• Vector search performance and infrastructure ([1], [2])

Talks and YouTube videos around AI and agents.

A selection of contributions:

• Vector search & workflow engineering - Qdrant Vector Space Day (Berlin, Sept. 2025)
• Anatomy of AI agents - Voxel51 Meetup (Stuttgart, April 2026)
• LiteParse: Local Document Parsing for AI Agents
• Build Agents from your Files (with LlamaAgents Builder)
• Qdrant talks & presentations

Building in public.

~5000 LinkedIn followers

900+ X followers

~500 GitHub followers

2.2K+ GitHub stars across projects

Active through blogs, YouTube tutorials, OSS demos and technical talks.

Rust • S3-backed, git-versioned object storage for AI agents • 79 stars

Aggit is git designed for your AI agents.

Add atomic changes, commit them, back them up to a remote S3 server, without never polluting your branch history, and push to GitHub only when you’re satisfied with the result.

Rust • Convert everything to PDF • 249 stars • 20K+ downloads

Package and CLI for converting heterogeneous formats into PDF workflows for GenAI applications.

Adopted by developers building document pipelines.

Rust core • CSV data loader for TypeScript, inspired by Polars and Pandas

Sunbears is a native TypeScript library, cross-compiled from Rust.

Able to read a 1.000.000 lines CSV in 0.3s in Node, SunBears is designed to bear the weight of data-intensive workloads for TypeScript backends.

Rust • Unified interface to OpenAI and Anthropic compatible LLM APIs

Call OpenAI and Anthropic compatible LLM APIs through a single interface, from Rust and other languages.

Rust • Make-like task executor for Unix

A Make-like task executor for Unix operating systems, written in Rust.

Generalist AI agent supporting ACP, MCP and sandboxed execution

Agent framework with support for filesystem tools, memory, TODO tracking, AgentFS and multiple model providers.

Safer coding agents through filesystem sandboxing

Run Claude Code/Codex inside AgentFS rather than real filesystems to reduce destructive operations.

Spotify Wrapped for GitHub

Full-stack Go application with authentication, caching, monitoring and analytics.

Performance, scalability and infrastructure for search

• Developer Relations focused on performance, scalability and infrastructure for search
• Authored production-ops articles on the Qdrant blog: Memory Tiers in Qdrant (pinned/cached/cold tiers × quantization, benchmark-backed) and Configure Qdrant’s Optimizer for Predictable Search Latency
• Presenting Qdrant talks: qdrant-talks.clelia.dev

Building AI infrastructure and agentic developer tooling

• Working on LlamaAgents and LlamaParse Platform (agent capabilities, knowledge base/agentic retrieval capabilities)
• Maintaining open source projects such as LiteParse and Semtools
• Producing demo for LlamaParse Platform products, also in partnership with other companies such as Google, Render, Qdrant, LanceDB, Auth0
• Producing tech content such as blogs and videos, and engaging with the community at conferences and meetups

Legal document understanding with AI

Contributed to pipelines processing legal documents through AI agent orchestration.

Genomics & quantitative biology

Researched human-driven impacts on Drosophila population genetics using genome data.

AI for architecture & design workflows

Early engineering work around AI-powered products for architects and designers.

Build a Smart Financial Assistant with LlamaParse and Gemini 3.1

Co-authored technical article on the Google Developers Blog demonstrating how to combine LlamaParse with Gemini 3.1 to build AI-powered document workflows.

Building Document Pipelines That Actually Scale

Contributed a deep-dive on architecting scalable document ingestion pipelines using LlamaParse, covering deployment patterns and performance considerations on Render’s infrastructure.

Securing AI Document Pipelines with LlamaIndex and Auth0

Partnered with Auth0 to write a guide on adding authentication and authorization to AI-powered document workflows, integrating the LlamaParse Platform services with Auth0’s identity platform.

Smart Parsing Meets Sharp Retrieval — Combining LiteParse and LanceDB

Collaborated with the LanceDB team to demonstrate how LiteParse and LanceDB complement each other for efficient multimodal document parsing and vector retrieval at scale.