How a Mobile Mapping System Stores 1 GB of Survey Data Using Only 108 KB in Postgres

·2 views

A mobile mapping survey using a LiDAR scanner and panoramic camera generates just over one gigabyte of data per run, but only a 108 KB trajectory file — recording the vehicle's path — is stored in the database. The remaining heavy files, including 507 MB point clouds and 566 MB spherical photos, are kept on disk and referenced via file paths in the database model. This approach keeps a database covering 2.7 million road features across 100 layers at just 2.3 GB total, even as raw survey data grows rapidly. Point cloud viewing in the browser relies on an octree structure of small files fetched on demand, a workload better suited to a static file server like nginx than a database connection pool. The core design principle is storing only metadata, geometry, and file pointers in Postgres — leaving binary data where it is most efficiently accessed.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

How Actian VectorAI DB Works Best When You Keep Embedding Separate

A developer building a local knowledge base with Actian VectorAI DB found that the real challenge was correctly separating responsibilities across tools. The embedding model (all-MiniLM-L6-v2) converts text into 384-dimension float vectors, while VectorAI DB solely handles vector storage and nearest-neighbor search — it never processes raw text. Automation tool n8n was kept strictly as a trigger layer, connecting inputs like Slack or webhooks to a search API without owning any core logic. Tests on a six-note corpus, including a deliberate off-topic distractor, confirmed that semantically relevant queries consistently returned correct top results using cosine similarity scores. The project, published on GitHub, demonstrates that clearly defining each tool's role is key to avoiding common pitfalls like single-entry test corpora or over-engineered workflow chains.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer builds CLI tool Serval to map the blast radius of code changes

A developer has created Serval, a local-first command-line tool designed to help engineers understand the potential impact of changes in large or unfamiliar codebases. The tool analyzes multiple signals — including dependency graphs, Git history, co-change frequency, and CI/CD configurations — to generate an evidence-based risk score for any targeted file. Serval currently supports repositories written in JavaScript, TypeScript, Go, Python, Java, and C, and can parse CI pipelines from GitHub Actions, GitLab CI, Azure Pipelines, and Jenkins. Users can also define critical-path rules via a .serval.yml file to flag high-priority areas such as authentication or payment logic. Rather than making AI-driven predictions, Serval aims to surface concrete, repository-level evidence so developers can make more informed decisions before merging changes.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Adversa AI Shows How Encrypted Prompts Bypass Grok and Gemini Safety Filters

Cybersecurity firm Adversa AI demonstrated a zero-click attack in which AES-encrypted malicious instructions embedded on a webpage bypassed the text-based safety filters of both Grok and Gemini AI assistants. Because the payload was ciphertext, no content filter flagged it as harmful during the input stage. When the AI models used their built-in code execution capabilities to decrypt the blob, the resulting plaintext was treated as trusted internal output rather than external untrusted content. This provenance loss allowed the decrypted instructions to direct the model to make an outbound request, exfiltrating the user's chat history, name, and location to an attacker-controlled server without any user interaction. The vulnerability highlights a fundamental trust-boundary flaw in AI agents that combine code execution with unsanitised web content, which standard pattern-matching guardrails cannot address.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer shares Claude Code setup that cuts repetitive AI corrections by 80%

A developer writing for DEV Community described how they transformed Claude Code from a basic code-snippet tool into a functional pair programmer by configuring context files and automation hooks over a weekend. The key change was adding a concise CLAUDE.md file in the repository root, which outlined the tech stack, package manager rules, and coding conventions, eliminating roughly 80% of trivial errors on day one. They also created dedicated rule files for recurring mistakes and reusable 'skill' prompts for repetitive tasks like PR reviews and commit message formatting. Automated hooks were set up to run tests and formatters after each file edit, forcing the AI to immediately address any failures. The author cautioned against overbuilding hooks, noting that chaining too many checks at once created more friction than it resolved.

0 comments Read more at DEV Community