6 min readDevelopmentFeatured
Continued Pretraining for Domain Adaptation
A technical overview of continued pretraining strategies for domain adaptation, covering corpus mixing, CPT, and drift mitigation.
All tags
Tag
Nlp
Read 13 articles on Nlp by Ashish Srivastava — practical guides covering implementation patterns, security, and production lessons.
13 posts
6 min readDevelopment
HyDE and Query Expansion: Do They Still Help?
We test HyDE and query expansion against current baselines to see if these methods still provide a significant boost to retrieval accuracy.
6 min readDevelopmentFeatured
Query Rewriting: The Underrated RAG Upgrade
Query rewriting is a critical yet often overlooked step in RAG pipelines. Learn how to boost recall and handle conversational context effectively.
Part 9 · RAG from Zero to Production
6 min readtechnicalFeatured
Chunking Complex PDFs: Tables, Columns & Scans
Learn how to handle difficult PDF structures like tables, multi-column layouts, and scanned documents during the chunking process.
6 min readDevelopment
Fine-Tuning an Embedding Model for Domain Retrieval
Learn how to adapt pre-trained sentence transformers to specific domains by leveraging hard negatives and contrastive loss functions for better retrieval accuracy.
6 min readDevelopmentFeatured
Tokenization: BPE, Unigram, and Why It Breaks
Explore how BPE and Unigram tokenization work in practice, why SentencePiece is popular, and the specific failure modes like glitch tokens that developers need to watch for.
Part 6 · Transformers Under the Hood
6 min readDevelopment
Positional Encodings: Sinusoidal, RoPE, and ALiBi
Explore how Sinusoidal, RoPE, and ALiBi positional encodings enable transformers to handle long contexts and extend sequence lengths effectively.
Part 3 · Transformers Under the Hood
6 min readDevelopmentFeatured
The Lost-in-the-Middle Problem and How to Route Around It
Large language models often struggle to retrieve information placed in the middle of long contexts. This article explores position bias and effective workarounds.
Part 2 · Context Engineering
6 min readDevelopmentFeatured
Hyperparameters That Actually Matter
Focus on the hyperparameters that drive model performance: learning rate, batch size, and epochs. A guide for ML engineers.
Part 6 · Fine-Tuning LLMs Hands-On
6 min readDevelopmentFeatured
Reranking: The Cheapest Quality Win in RAG
Reranking is the most cost-effective way to boost RAG accuracy. Learn how to implement cross-encoders like Cohere and BGE for better retrieval results.
Part 6 · RAG from Zero to Production
6 min readDevelopmentFeatured
Choosing an Embedding Model in 2026
Navigate the 2026 landscape of embedding models. Learn how to evaluate MTEB scores, leverage Matryoshka dimensions, and handle multilingual data for production RAG.
Part 3 · RAG from Zero to Production
6 min readtechnicalFeatured
Chunking Strategies That Survive Real Documents
Real documents are messy. Learn how to implement effective chunking strategies, semantic splitting, and overlap techniques to improve RAG retrieval accuracy.
Part 2 · RAG from Zero to Production
6 min readDevelopmentFeatured
Sampling Parameters Demystified: Temperature, Top-p, and Top-k
A technical breakdown of sampling parameters like temperature, top-p, top-k, and min-p, explaining how they control the randomness and quality of LLM generation.
Part 7 · LLM Inference from Scratch