SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World
Back to feed
Programming

Sahayak: Open-Source AI Legal Assistant Built to Admit Its Own Limits

0
·1 views

A developer built Sahayak, a RAG-based AI legal assistant, as part of the Hack2Skill PromptWars hackathon to make legal information more accessible in plain language. The tool lets users ask questions about tenancy, consumer rights, or contracts, upload legal PDFs for summarization, or speak queries via voice transcription powered by Groq Whisper. Unlike typical LLM chatbots, Sahayak explicitly states when it lacks sufficient context rather than generating confident but potentially incorrect answers. The system uses ChromaDB for vector retrieval, sentence-transformers for embeddings, and a FastAPI backend, with security measures including prompt-injection resistance, file validation, rate limiting, and automated CI workflows. The project is MIT-licensed and publicly available on GitHub, with planned improvements including multi-language support and persisted chat sessions.

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)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Laya offers a fast, local 322M-parameter alternative to LLM-based text classification

Laya is an open-source, non-autoregressive decision engine designed to replace costly large language model calls for routine classification tasks such as routing, moderation, and escalation. Unlike standard LLM-as-a-judge setups, Laya runs a single encoder forward pass to answer typed questions — including label selection, scoring, and yes/no probabilities — without generating any tokens. The project ships three model checkpoints on Hugging Face, including a 322M-parameter multilingual model covering over 100 languages, and reports inference speeds as fast as 7.2ms per question when batched on a T4 GPU. Laya runs entirely locally with no API keys or GPU required, needing roughly 1.5GB of disk space and 1GB of RAM. Released on September 18, 2025, the project accumulated over 26,000 GitHub stars within nine days, reflecting widespread frustration with the latency and cost of using frontier models for simple classification decisions.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

pixi-tiledmap lets developers edit and export Tiled maps directly in the browser

A TypeScript library called pixi-tiledmap has been developed to load, render, edit, and export Tiled maps using PixiJS v8 in the browser. The library allows developers to place tiles programmatically via a setTile API that accepts layer names, grid coordinates, and tileset references, updating both the visual display and underlying map data in real time. Edited maps can be exported as standard Tiled JSON (.tmj) files using the exportMap function, making them immediately reopenable in the Tiled map editor. Because Tiled stores tileset images as file paths rather than embedding them, the library also supports bundling the map and its assets together into a single ZIP file using fflate. The workflow enables a full round-trip edit cycle — from Tiled to browser and back to Tiled — without leaving the standard .tmj format.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer releases fzgrep, a parallelized fuzzy text matcher for Unix pipelines

A developer has released fzgrep, an open-source command-line tool written in pure C that performs fuzzy line matching within standard Unix pipelines. Unlike grep, which requires exact matches, fzgrep tolerates typos by computing Levenshtein distance using a memory-efficient single-row cache. The tool uses OpenMP-based parallelism through a chunk-based MapReduce model, distributing distance calculations across multiple CPU cores. Additional features include word-level matching, similarity scoring, and coordinate tracking for matched tokens. The project is available under the GPL-2.0 license on GitHub, and the developer is seeking community feedback on handling large-scale streams and potential algorithmic improvements.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Ukrainian CyberSec trainer shares performance-first self-learning system for tech skills

Ivan, a cybersecurity instructor and blogger at White2Hack, has published a practical self-learning framework drawn from years of teaching and personal study in IT security. The system, which he calls performance-first learning, centers on setting verifiable, action-based goals rather than vague intentions like 'learn API security' or 'watch a DevSecOps course.' Ivan argues that knowing many learning techniques — such as spaced repetition or the Pomodoro method — does not guarantee real retention, as students often struggle to explain topics without prompts. His core study cycle moves from defining a measurable outcome through mapping a subject, working through examples, self-explaining, and retrieving information without aids, before applying knowledge to new tasks. He notes the approach applies beyond cybersecurity, covering areas like English language learning, technical reading, and interview preparation.

0 comments Read more at DEV Community
Sahayak: Open-Source AI Legal Assistant Built to Admit Its Own Limits · ShortSingh