Case study · 2024 — Present

AI-Driven Workflow Engine

RAG + LLM orchestration layer that turns repetitive ops work into automated, reviewable decisions.

Role: Backend / AI Engineer
Company: NEWPRODATA
Year: 2024 — Present
Status: Shipped

01 — Problem

Teams were losing hours to repetitive triage, data enrichment, and document-writing tasks. Off-the-shelf LLM tooling either hallucinated or couldn't ground on internal data.

02 — Approach

Built a Django-native workflow engine around a RAG pipeline: chunk + embed internal documents into a vector store, retrieve top-k on each prompt, hand the LLM grounded context with explicit JSON schemas. Routed between Claude, ChatGPT, and Bedrock per task cost/latency. Added an audit log so every AI decision is reviewable. Structured output validation before anything writes back to the DB.

03 — Outcome

Replaced multiple manual workflows while keeping humans in the loop where it matters. Cut average task turnaround materially while preserving traceability.

◆
Grounded answers — no orphan hallucinations in production
◆
Full audit trail for every AI-generated artifact
◆
Multi-model routing (Claude / ChatGPT / Bedrock)
◆
Retrieval latency under 200ms p95

04 — Tech stack

Python Django AWS Bedrock Claude ChatGPT Vector DB RAG FAISS PostgreSQL Celery

More work

2024 — Present

Datasprings — Data Automation & API Platform

Django + Neo4j employment platform with scraping pipelines, RAG-based search, and cloud-native deployment.

Read the case

2024 — 2025

Timepaq — Attendance & Employee Management

Web-based attendance and HR system with biometric-device integration. Led design, delivery, and team coordination.

Read the case