💡 Inspiration

Procurement and legal departments in organizations waste weeks manually auditing massive tender documents, checking compliance policies, and validating Bill of Quantities (BOQ) arithmetic. Errors in these audits cost millions. We built TendAI to automate this cognitive burden, providing a localized, security-first SaaS system that leverages advanced agentic workflows and LLM reasoning.

⚙️ What it does

TendAI is a multi-tenant platform featuring:

  • Intelligent Next.js Workspace: A secure interface where users chat with an agent to analyze contracts, use @ mentions to link documents, and / commands for custom skills.
  • Multi-Agent Orchestration: A LangGraph state-machine that routes queries, retrieves chunks, performs risk audits, checks output quality, and verifies citations.
  • Knowledge Base Ingestion: A management portal for uploading permanent reference policies and templates, parsed using layout-aware engines.
  • Human-in-the-Loop Gate: Pauses autonomous executions for critical input or clarification from human experts before proceeding.
  • Grounding & Provenance: Generates answers that explicitly cite retrieved chunk IDs, validating source evidence to prevent hallucinations.

🛠️ How we built it & How Codex Accelerated Us

  • Codex & GPT-5.6: Codex was our primary engineering companion. It accelerated building the PostgreSQL multi-tenant repository isolation and helped write the state checkpointers for our Human-in-the-Loop flows in record time.
  • Backend: FastAPI (Backend API on port 8080) and LangGraph (Agent Orchestration).
  • Frontend: Next.js client featuring interactive mention-menus and markdown rendering with citational drawers.
  • Vector Database: ChromaDB for tenant-scoped vector search.

🧠 What we learned

  • Working with complex document layouts requires deep parsing pipelines (like Docling) rather than raw text extraction to preserve tables and hierarchies.
  • Designing multi-agent systems with explicit loops and checkpointers is more reliable than single prompt prompts.

🚀 What's next for TendAI

  • Deploying our API to production servers.
  • Providing offline/edge adapters for localized enterprise hosting (Local-first edition).

Built With

Share this project:

Updates