devix.
All work

AI & Automation

Devix AI Tender & Vendor Evaluation

A fully local LLM pipeline that scores vendor proposals against RFP requirements with cited evidence, weighted rubrics, and anomaly detection.

Client
Government & enterprise procurement
Completed
September 2025
Technologies
FastAPIPythonStreamlitOllamaPostgreSQLQdrantDocker

An open, four-stage tender evaluation platform that automates vendor scoring using local AI. It extracts requirements from an RFP, builds an evidence library from vendor submissions, scores each vendor against weighted criteria with citations, and flags anomalies — all running locally via Ollama so sensitive procurement data never leaves the organization. Bilingual Arabic/English UI with a Devix AI Guard layer for identity and output validation.

4

Pipeline stages

Local (Ollama)

Inference

Qdrant

Vectors

AR + EN

Languages

The challenge

Procurement teams spent weeks manually reading and scoring vendor proposals against RFP criteria — slow, inconsistent, and hard to defend. Government buyers also could not send confidential bids to cloud AI services.

The solution

A local pipeline: requirement extraction → evidence library (retrieval) → weighted vendor scoring with cited evidence → anomaly detection. Ollama runs the models on-prem, Qdrant powers retrieval, and Devix AI Guard validates identity and outputs — with a Streamlit operator UI and defensible, evidence-backed scores.

Architecture

FastAPI services orchestrate the stages, Qdrant stores vector embeddings, Ollama serves local LLMs, PostgreSQL persists results, and Streamlit provides the operator interface — packaged with Docker Compose for on-prem deployment.

Key features

  • Automatic RFP requirement extraction
  • Evidence library with retrieval
  • Weighted vendor scoring with citations
  • Anomaly and inconsistency detection
  • Devix AI Guard identity/output validation
  • Arabic and English operator UI
  • 100% local inference via Ollama
  • Docker Compose deployment

Have a similar problem to solve?

Tell us what you're building — we'll tell you honestly whether we're the right team for it.

Start a conversation