LLM · Legal Assistant2024

TRSKHOOB

An AI-native legal assistant that turns messy natural-language questions into precise, source-grounded consultation flows — in Persian.

Senior AI Engineer & Project Manager

trskhoob.ir
Loading preview
TRSKHOOB live preview

Timeline

2024 — ongoing

Team

AI lead + backend + product design

Role

Senior AI Engineer & Project Manager

Overview

TRSKHOOB is a domain-specialised LLM product for the Iranian legal market. It gives citizens and junior practitioners a guided consultation surface — voice or text — that translates real questions into structured legal reasoning, jurisdiction-aware answers, and next-step actions.

The problem

Legal help in Persian is expensive, slow, and hard to trust. Existing chatbots hallucinate, ignore local codes, and lose the thread across a multi-turn consultation. Users needed something that behaved like a careful junior lawyer — not an autocomplete.

Approach

I led product architecture, prompt engineering strategy, and the backend that stitches retrieval, orchestration and voice pipelines together. The system routes each turn through intent classification, jurisdiction-aware retrieval over curated legal corpora, and a structured response contract that separates facts, reasoning and disclaimers.

What I built

Key capabilities.

  • 01Multi-turn consultation memory with jurisdiction and case-type routing
  • 02Retrieval over curated Persian legal corpora with citation-aware answers
  • 03Voice-in / voice-out pipeline for low-literacy accessibility
  • 04Prompt orchestration layer with per-domain safety and refusal policies
  • 05Operator dashboard for reviewing sessions and improving prompts in loop

Outcomes

What shipped.

Consultation flows

20+

structured intents

Response contract

3-part

facts · reasoning · disclaimer

Modality

Voice + Text

Persian-first

This project

Live and in production.

Open the live project or start a conversation about collaboration.