Skip to main content
Ubaith SherifAI Engineer
AboutWorkExperienceStackResearchContact
Resume ↗
← Back to Projects
Enterprise AI · Founder Project

Unscript One

AI-Native Enterprise Workspace — RAG, Agents, MCP, Guardrails, and LLM Evaluation unified.

Unscript One is an enterprise-grade AI workspace that unifies Retrieval-Augmented Generation, AI Agents, Model Context Protocol, Guardrails, and LLM Evaluation into a single production platform for document intelligence, repository analysis, research automation, and enterprise knowledge management.

GitHub Repository ↗Live Demo ↗

Problem Statement

Organizations building AI internally repeatedly hit the same wall: chatbots that hallucinate, no visibility into whether answers are grounded, no control over tool access, and no way to measure quality over time.

System Solution

Unscript One delivers the core of the real enterprise AI problem: grounded answers via hybrid RAG, controlled tool access through MCP, measurable quality via automated Ragas evaluation, and full observability on one dashboard.


Architecture & Data Flow

01.

Next.js streaming workspace with real-time AI responses, inline citations, and multi-session chat history.

02.

FastAPI backend services for document ingestion, retrieval, agent orchestration, and guardrail enforcement.

03.

LangGraph-powered Research and Coding Agents with GitHub MCP integration for repository-aware workflows.

04.

Hybrid retrieval pipeline: BM25 + vector search with reranking, backed by Qdrant and Supabase PostgreSQL.

05.

Pre-generation guardrail layer for prompt injection detection, PII filtering, and request validation.

06.

Automated Ragas evaluation loop running on every deploy, with metrics surfaced on a public /status page.


Key Technical Features

  • 01.Enterprise RAG with recursive chunking, hybrid retrieval, reranking, and citation-based responses.
  • 02.AI Research and Coding Agents powered by LangGraph with real GitHub MCP integration.
  • 03.Prompt injection and PII guardrails running pre-generation with visible pass/fail badges.
  • 04.Automated LLM evaluation using Ragas with faithfulness and answer relevance metrics.
  • 05.AI Observability Dashboard and public Service Health Monitoring at /status.
  • 06.Multi-session AI conversations with document ingestion and knowledge management.

Engineering Challenges

  • 01.Building a production-ready hybrid retrieval pipeline balancing BM25 recall with vector search precision.
  • 02.Designing LangGraph agent flows that expose tool call traces transparently in the UI.
  • 03.Running Ragas evaluation automatically on every deploy without blocking the release pipeline.
  • 04.Keeping prompt injection and PII guardrails fast enough to run pre-generation without noticeable latency.

Complete Technology Stack

PythonFastAPINext.jsTypeScriptLangGraphGoogle GeminiQdrantPostgreSQLSupabaseMCPRagasTailwind CSSDocker

Related Projects

Enterprise Analytics

InsightAI Enterprise

Multi-Agent Decision Intelligence Platform with real-time analytics, semantic search, and AI-powered business insights.

Business Operations

Multi-Agent Business Automation

Human-reviewed automation for sales, support, and reporting with approval gates and full auditability.

Ubaith Sherif

AI Engineer · Full-Stack Developer

Building AI-native systems, retrieval pipelines, and production software.

GitHub ↗LinkedIn ↗Email ↗

Navigation

  • About
  • Projects
  • Experience
  • Stack
  • Research
  • FAQ
  • Contact

Legal

  • Privacy Policy
  • Terms of Service

© 2026 Ubaith Sherif · Built with Next.js & TypeScript

Coimbatore, India