Scaling Implicit Recommendations: ALS and Annoy in Production
I remember a time when our explicit rating system was barely limping along, struggling to provide meaningful recommendations because, let's b…
Latest posts
Unlocking Hidden Value: Building a Resilient Pipeline for Nepal Rastra Bank PDF Data Extraction
Have you ever spent hours manually sifting through Nepal Rastra Bank's (NRB) Quarterly Economic Bulletins, painstakingly copying numbers from…
Production-Ready FastAPI: Aggregating PyPI Stats with Pydantic, Async, and Smart Caching
I've spent countless hours building APIs that serve data from various external sources. It’s a common scenario: you need to expose curated da…
Beyond Simple Categories: Building a Resilient NLP Pipeline for Unstructured Financial Data
Have you ever found yourself staring at a wall of unstructured text – perhaps a stream of API logs, a news feed, or even your own bank statements…
Quantizing DistilBERT for Real-time Edge Inference: A Practical Guide to TinyML Text Classification
Master the art of quantizing small, pre-trained transformer models (like DistilBERT) to deploy efficient, real-time text classification on resour…
From Cloud to Kathmandu: Architecting Quantized LLM Inference on the Edge for Nepali Businesses
Have you ever felt the friction between the immense potential of large language models and the stark realities of deploying them in resource-cons…
Beyond `dict.get()`: Building a Resilient Data Cleaning Pipeline for Inconsistent API Streams
Leverage Pydantic for declarative schema validation and robust type coercion, combined with Pandas for efficient transformation, to construct a r…
From Generalist to Specialist: Fine-Tuning LLMs with LoRA and QLoRA for Production Efficiency
Remember that exhilarating feeling when we finally wrangled our LLM inference costs, deploying efficient batching and quantization strategies? (I…
Bridging the Gap: Mitigating Training-Serving Skew with a Production-Ready Feature Store
Learn to implement a lightweight feature store using Feast to consistently define, compute, and serve machine learning features, thereby eliminat…
Beyond Tokens: Crafting a Cost-Optimized LLM Inference Layer with Batching, Caching, and Quantization
Master the implementation of dynamic batching and semantic caching, complemented by model quantization, to significantly reduce LLM inference cos…