Skip to content
Sai Eswar Gujjala
  • Professional work
  • Deployed internally

Intelligent Procurement Copilot

An LLM copilot that turns a procurement request into a vendor recommendation report. It reads the request, finds and ranks vendors with semantic search, checks compliance, and writes up the recommendation.

My role
End-to-end owner.
Key result
The procurement cycle went from about 3 days to under 30 minutes per request, compared with the manual process.
Technologies
  • Python
  • LangGraph
  • Pinecone
  • RAG
  • LLM tool use
Skills shown here

What I owned

  1. Problem framing
  2. Data preparation
  3. Modeling
  4. Evaluation
  5. Application/API
  6. Deployment

Built at Zetwerk for internal use. Described in simplified form; no customer or supplier names, data, code or internal links are shown.

In [1]:Problem and constraints

For each procurement request, someone had to read it, search for suitable vendors, check compliance and write a recommendation. That took about 3 days per request.

In [2]:Data and preparation

Procurement requests, vendor profiles and history, and compliance rules and documents.

In [3]:Baseline and evaluation design

The time from request to recommendation report, compared with the manual process.

In [4]:Architecture and modeling decisions

A LangGraph agent with explicit steps:

  1. parse the request into structured requirements;
  2. search vendors semantically in Pinecone;
  3. rank them;
  4. check compliance through tools;
  5. write the report.

Separate steps keep each stage checkable, and tools enforce the compliance rules instead of the model's memory.

In [5]:Implementation and integration

Vendor information is indexed in Pinecone for retrieval. The agent calls tools for compliance checks and assembles a report with the ranked vendors and the reasons for each.

In [6]:Deployment and operation

Status: Deployed internally

Used internally at Zetwerk.

Out[6]:Results and evidence

  • The procurement cycle went from about 3 days to under 30 minutes per request, compared with the manual process.

In [7]:Limitations, failures, and lessons

  • Recommendations are only as current as the vendor data, compliance rules need upkeep, and the final choice stays with the procurement team.