Focused on
Adoption
AI & legal-tech transformation

From legal data to adopted tools. Based in Tallinn, Estonia.

See the case studies

Awards

  • Winner — Mirka M.INC Innovation Challenge (Finland)

  • Winner — Base44 Hack4Good hackathon

Sagar Yadav
AI Lead, Systems Development — Government of Åland, since July 2024.
From legal data to adopted tools
Based in Tallinn, Estonia
Legal-tech work
Stack
PythonRustGoSQLiteDockerAWSCloudflareOpenRouter
/b4.svg
Fluent, native, learning
English, Hindi, Estonian
AI legal advisor for Estonian and EU law.
Est Law is live
Tallinn
Get in touch

Case studies

Problem, approach, governance, and what came out of it.

01

Est Law

AI legal advisor for Estonian and EU law.

Live site
  • Python
  • SQLite
  • OpenRouter
  • Cloudflare

Problem

Answers have to stay inside Estonian law and EU law, and those sources can conflict.

Outcome

Live at estlaw.stepinbooking.com. Answers are planned in Estonian, English, and Russian.

Approach

  • Structured Estonian law from Riigi Teataja, plus 33,114 EU acts from EUR-Lex.
  • A source hierarchy resolves conflicts: EU law → Constitution → treaties → acts → regulations.
  • Agentic tree-RAG: act cards → section maps → provision summaries → full text.
  • Daily update pipeline: sha256 diff, only changed acts re-queued, stable IDs.
  • Answers planned in Estonian, English, and Russian.

Governance / QA

  • Bulk summarisation split into 1,653 validated, hash-checked work units.
  • Coding agents ran those units across several machines, with progress tracking, validators, and cost caps.

02

ÅlexGPT / Law Navigator

AI legal search for the Government of Åland, with Jacob Pesämaa. July 2024–present.

  • Go
  • Rust
  • OpenAI
  • Claude
  • OpenRouter
  • KeyDB
  • Docker
  • AWS

Problem

The statute archive had to be digitised before it could be searched, and bad file naming had silently dropped documents from the government dataset.

Outcome

Private repositories. No public link.

Approach

  • Digitised the statute archive: Go downloader, Rust PDF-to-image, OCR enhancement, and human-in-the-loop review.
  • Proposed a naming standard and a data-readiness review after the dropped documents came to light.
  • RAG chatbot using OpenAI text-embedding-3-large and Claude via OpenRouter, with a KeyDB cache, Docker, and AWS.
  • Swedish-to-English law-XML translation, with highlighted source passages.
  • Successor: embedding-free agentic RAG over 5,236 in-force Finnish laws.

Governance / QA

  • Human-in-the-loop review of the digitised archive.
  • A naming standard and a data-readiness review for the government dataset.

03

Law Text Editor

Bilingual English and Swedish editor for drafting legislation.

  • XML

Problem

Drafting legislation in two languages needs structure, amendment tracking, and a controlled path from draft to approved text.

Outcome

No public link.

Approach

  • Structured editing for metadata, chapters, and paragraphs, with amendment tracking and validation.
  • Live XML preview and export.
  • An approved-versus-draft workflow with four-eyes (two-person) approval.

Governance / QA

  • Validation on the draft.
  • Four-eyes approval: two people sign off before a draft is treated as approved.

04

AI Electronics

AI that detects PCB defects and gives repair guidance. Co-founded with Jacob Pesämaa, 2024–present.

Live site

Problem

Detect defects on printed circuit boards and give repair guidance.

Outcome

Winner — Mirka M.INC Innovation Challenge (Finland). Live at aielectronics.vercel.app. The funding pitch was for up to €20k in the first phase.

Approach

  • Took it through problem validation, proof of value, and customer interviews.
  • Pitched for up to €20k in first-phase funding.

Governance / QA

  • Problem validation, proof of value, and customer interviews before the funding pitch.

How I run AI transformation

Discovery, then a pilot, then rollout.

  1. 01

    Discovery

    • Start from the workflow and the data it depends on.
    • Set a data-quality standard before building. On the Åland statute archive, bad file naming had silently dropped documents from the government dataset. The response was a naming standard and a data-readiness review.
  2. 02

    Pilot

    • Break the work into validated units with clear ownership.
    • Put validators and QA gates on those units. Est Law’s bulk summarisation ran as 1,653 validated, hash-checked work units, with progress tracking.
    • Set cost caps, and require approval before paid usage.
    • Keep a human in the loop. The Åland statute archive was reviewed by a person after OCR.
  3. 03

    Rollout

    • Stage the path from beta to production. Tallinn grocery deals moved from beta to production.
    • Track progress and benefit.
    • Keep sources current. Est Law’s daily pipeline uses a sha256 diff, re-queues only changed acts, and keeps stable IDs.

Other builds

Experience

  1. Jul 2024–present

    AI Lead – Systems Development

    Government of Åland

    AI legal search with Jacob Pesämaa (ÅlexGPT / Law Navigator): statute-archive digitisation, a naming standard, and a data-readiness review.

  2. 2024–present

    Co-founder / CTO

    AI Electronics

    AI that detects PCB defects and gives repair guidance, co-founded with Jacob Pesämaa. Winner — Mirka M.INC Innovation Challenge (Finland).

  3. 2022–2024

    Freelance full-stack engineer

  4. 2021–2022

    Co-founder

    BerryBot

    Computer vision for strawberry-picking robots. Second place in a hackathon.

  5. Since 2007

    Founder

    Earlier ventures

    Entrepreneurial track record: Biotex Blockchain, Brandeligion News, Diplomat Auctions, and idHUG.com.

Education and certifications

  • 2007–2010

    MBA, Amity Business School

    Lean, Six Sigma Green Belt, Kaizen, and ISO 9001 training.

  • 2022–2023

    Turing College

    Data Science / ML.

  • 2022–2024

    grit:lab

    Software engineering: AI, cyber security, and DevOps. Top of cohort, with a GPA double that of the second-ranked student.

  • Highest GPA in cohort at Kood Tech (01-edu network)

  • WIPO Intellectual Property Rights certification

Languages

  • English · Fluent
  • Hindi · Native
  • Estonian · Learning

Contact

Tallinn, Estonia

Drag the piece into the slot

Mouse, touch, or keyboard: focus the piece and press Enter.