DataDrill logo

DataDrill

enhanced
Akta

Akta - AI Legal Intelligence Platform

Akta

Pre Seed to Seed Startup
Company Type
Novi Sad, Serbia
Location
Project work
Engagement Model
1 - 5 people
Team Size
3 - 6 Months
Duration
$21K - $50K
Budget

About

Akta is a production-grade B2B SaaS AI legal assistant designed for the Serbian legal market, addressing malpractice risks by automatically detecting statute amendments and providing source-cited answers in standard Serbian legal citation format.

Challenge, approach, and impact

General-purpose AI models hallucinate legal content

The entire architecture had to be rebuilt around grounding: the model is explicitly prohibited from drawing on its training knowledge of law and must rely solely on retrieved, verified source texts.

Precedents cite laws that have since been amended

Akta had to detect and surface per-article statute changes between the date of a cited decision and today to address the risk of lawyers unknowingly referencing repealed or amended provisions.

Suitable embedding model for Serbian legal language

A proprietary retrieval model had to be built and fine-tuned from scratch on an in-house legal corpus due to the lack of suitable embedding models for Serbian legal language.

Machine-readable legal corpus

Approximately 50,000 court decisions had to be acquired, processed, enriched, chunked, and indexed into a vector database to create a machine-readable legal corpus.

Article-level retrieval for statutes

The system needed surgical article-level retrieval to identify relevant article numbers and extract only those articles (plus adjacent context) for inclusion.

Live progress updates for AI pipelines

The system needed to stream live progress updates at each named pipeline stage to reduce anxiety and abandonment due to long AI pipelines.

Prompt injection defenses for untrusted documents

Prompt injection defenses had to be built into every ingestion path to treat content uploaded by lawyers as untrusted evidence, not as instructions for the model to follow.

AI cost control in a flat-fee B2B market

Tiered credit economics were engineered to ensure heavier operations consumed more credits than lighter ones, making the unit economics viable at scale in a flat-fee B2B market.

How we built

SaaS
Learning Management Systems (LMS)
Conversational AI
Web Apps
CRM

Testimonials

Marko Nikolic, AI Systems Architect @ DataDrill

DataDrill

Verified Testimonial

Marko emphasized the importance of trust in AI solutions for legal professionals. He highlighted the significance of providing verifiable answers grounded in retrieved sources, inline citation linking, and amendment detection to ensure lawyers can check the information provided.

Team structure

Client team

Danijel's avatar

Danijel

Project stakeholder

Project stakeholder

The client stakeholders at Akta were working closely with the team at DataDrill

Agency team

3 x Team's avatar

3 x Team

Production

Similar Case Studies

Slide 1 of 0
DataDrill's logo

Chat with agency

Let's turn your ideas into reality—connect with our team.