CTO & Co-Founder · 2022 to present

DigiLawyer

DigiLawyer is a legal intelligence platform for Pakistani law. It brings judgment and statute search, citation intelligence, document analysis, AI-assisted research, and drafting into one workspace.

Visit DigiLawyer ↗

My role

I lead the technical direction of the product. My work covers system architecture, agents, retrieval, infrastructure, data and access boundaries, evaluation, engineering sequencing, and production reliability.

The product itself is shared work across engineering, legal validation, corpus work, design, operations, sales, and institutional rollout.

The problem

Pakistan's legal record is spread across judgments, statutes, journals, databases, and PDFs that are difficult to search together. Useful legal work has to find the right authority, keep its source attached, respect access boundaries, and leave the final judgment with the lawyer.

Model memory is not enough. A fluent answer with the wrong authority is still wrong.

Decisions

Give different work different boundaries

Research, document analysis, drafting, review, and final synthesis need different tools, context, and failure handling. We keep those responsibilities explicit instead of giving one agent every capability.

Keep sources connected to the answer

Legal research works from retrieved sources, not model memory. Semantic and text retrieval work together with citations and case-treatment context so a lawyer can inspect the authority behind an answer.

Make access a server decision

Models only receive context a user is allowed to access. Permissions come from authenticated server state and are checked outside the model.

Treat long-running work as its own lifecycle

Agent runs do not fit neatly inside a normal application request. State and execution are separated by responsibility so progress, failure, recovery, and completion can be explicit product states.

Evaluate the path, not only the prose

A legal answer can look convincing while retrieval, routing, tool use, or citations are wrong. We evaluate those parts independently, then test the full workflow.

The tradeoff

One agent with every tool would be simpler to draw. It would be harder to reason about. We accept more explicit orchestration in exchange for clearer boundaries, independent evaluation, and tighter control over the context and tools each task receives.

Company evidence

DigiLawyer is live. The company reports 20,000+ users and 52 indexed legal journals. UET Lahore and Dawn reported in December 2025 that the platform was being used by government institutions, the judiciary, and law firms.

These are company outcomes. My contribution is the technical direction described above.