Achyar Ulul Amri

Medical Claims Document Intelligence Platform

Delivered an analytics platform that unlocked insights from hospital PDFs using OCR and LLMs at 95% extraction accuracy, with human-in-the-loop review feeding comprehensive medical-cost dashboards.

RAG-Powered Document Chatbot

Delivered a secure, RAG-powered document chatbot for instant fact-checking with 90% accuracy and automated citations linking directly to the exact source PDF page.

Retail Analytics Suite

Delivered a retail analytics suite across 4 markets (Indonesia, Hong Kong, Taiwan, Macau); cross-merchandising recommendations added ~5% revenue and replaced 3-4 day manual reporting with self-serve BI dashboards.

No-Code ML Development & Deployment Platform

Built a no-code end-to-end ML platform hosting 8 production models, authored an internal Python data-cleaning library, and ran company-wide AI/ML training.

Brand Intelligence Text-Analytics Suite

Built a BERT-based brand-perception product (sentiment, topic clustering, social-network analysis) over 4.8K news and Twitter records on Indonesian automotive brand keywords (2020-2024).

Palm-Oil Yield (FFB) Forecasting Model

Engineered a palm-oil yield (FFB) forecasting model at ~15% error rate (MAPE), enabling data-driven production, resource-allocation, and budget-planning decisions.

Reusable AI Product Catalog

Led a catalog of reusable AI products and tooling so a model built once could be deployed across several subsidiaries instead of being rebuilt each time.

Vehicle Price & Stock Aggregation Tools

Built a used-car price engine (regression on online-listing data) and a stock-aggregation tool, productized for automotive resale subsidiaries.

B2B & B2C Repurchase-Propensity Models

Delivered car & motorcycle repurchase-propensity models (B2B and B2C) that helped field sales find qualified leads.

Vehicle Repurchase & Recommendation Model

Delivered an ML-powered vehicle repurchase & recommendation model predicting buying timelines and next-model preferences, quadrupling conversion from 0.4% to 1.7%.

Customer Segmentation Models

Built customer-segmentation models (input, base, retention, outflow) to tailor strategy per segment.

Vehicle Demand-Forecasting Model

Built a vehicle-specific demand-forecasting model (FB Prophet), halving stock prediction error from 64% to 33%.

Fraud-Detection Model

Developed a fraud-detection model for a group mobility platform, preventing losses equivalent to 20% of the promotional budget.

Churn Prediction Model

Delivered churn prediction for after-sales branches, retaining more customers via high-potential churner lists for proactive treatment.