LEGAL AI & DOCUMENT INTELLIGENCE
Building a legal AI platform, end to end.
Judicio connects legal research, document review and drafting in one workspace. Rajkumar independently built its product architecture, application, AI workflows and production infrastructure.
- PROJECT
- Judicio
- RAJKUMAR’S ROLE
- Founder · Full-stack product development
- ENGAGEMENT
- January 2026 — present
The engineering challenge
A legal research workflow needs more than a generated answer. It must retrieve relevant material, preserve the evidence and let a professional inspect the sources. Document tasks also involve long-running processing that needs to recover from interruption without losing track of the work.
Research agents with evidence behind the answer
Rajkumar built legal research agents using PydanticAI. The workflow includes structured research planning, parallel tasks, evidence synthesis and citation verification. Responses are validated against schemas, and streamed progress makes longer tasks visible to the user.
Hybrid search for legal information
Judicio’s retrieval layer combines BM25 keyword search with vector similarity in Turbopuffer. Reciprocal rank fusion brings the results together, with optional reranking. This gives the system ways to find exact legal terms as well as material related to the meaning of a question.
Document processing as a durable workflow
Temporal coordinates OCR, text extraction, chunking, embeddings and enrichment. Retry and cancellation handling are part of that pipeline. AI-assisted review and drafting sit above the processed documents, alongside usage tracking and structured results.
An application designed through production
The application combines Python, FastAPI, asynchronous SQLAlchemy, PostgreSQL and Redis with React, Next.js and TypeScript. Deployment uses Google Cloud, containers and automated release tooling, with OpenTelemetry tracing and backend and frontend checks.
What was delivered
The result is a launched legal AI product connecting research and document work in a shared application. Source inspection and professional review are part of the workflow, alongside the generated output.
What this experience brings to your project
The transferable engineering pattern is the combination of useful retrieval, structured agent behavior, durable document processing and a complete product interface. Those capabilities can support other document-heavy business workflows.
Rajkumar’s full engineering experience