Work That Speaks for Itself

Real projects. Real results. Shipped by vetted crafters using AI tools — without cutting corners.

Fintech / Compliance

AI Features

The Challenge

Legal team manually reviewing compliance documents for deal closure — taking 6+ hours per deal, expensive, and prone to oversight.

The Approach

Built a RAG pipeline with LlamaIndex that ingests regulatory docs, vectorizes them, and retrieves context on-demand for an LLM to answer compliance questions in seconds.

The Result

93% time savings per deal. What took 6 hours now takes 25 minutes. System handles 150+ documents without retraining.

93%

time saved

6h → 25m

per deal

150+

docs indexed

PythonLlamaIndexOpenAIPineconeFastAPI
Duration8 weeks

E-commerce / Retail

Data Product

The Challenge

Recommendation engine using basic LLM calls couldn't scale past 1,000 users and was burning $50K/month on API costs.

The Approach

Redesigned with two-tower architecture (user embeddings + item embeddings), added real-time feature engineering with Kafka, and optimized with approximate nearest neighbor search.

The Result

Now serves 500K users with <100ms latency and 78% cost reduction. System generates $2M+ in incremental annual revenue.

500K

users

<100ms

latency

78%

cost reduction

PythonPyTorchKafkaRedisPostgreSQLKubernetes
Duration12 weeks

Manufacturing / QA

AI Features

The Challenge

Manual quality inspection catching only 78% of defects; inspectors working 8-hour shifts with fatigue-induced misses. No 24/7 coverage.

The Approach

Fine-tuned YOLOv8 on 5,000 product images with 12 defect types. Deployed on edge (Nvidia Jetson) for real-time inference. Integrated with factory line sensors and alert system.

The Result

97.3% accuracy, 24/7 monitoring, zero downtime integration. Reduced scrap rate by 34%, preventing $400K+ annual losses.

97.3%

accuracy

24/7

monitoring

$400K+

prevented loss

PythonYOLOv8Nvidia JetsonOpenCVTensorRT
Duration6 weeks

Stop Choosing Between Speed and Quality

Connect with a crafter who has shipped dozens of products and knows exactly how to use AI tools without creating a codebase you'll regret.