AI & Engineering · 11 min
Your Vector Database Is Lying to You About Multi-Hop
Dense retrieval hits a proven geometric ceiling on multi-hop questions and fails silently — similarity scores stay confident even when the evidence is missing.
Tag
#rag
12 posts
AI & Engineering · 7 min
A GenAI Project That Delivered Measurable Business Impact
A GenAI project succeeds or fails on whether you can name the business metric it moved — and prove it. The architecture is the price of entry, not the win.
AI & Engineering · 7 min
Data Governance for GenAI in Regulated Industries
In regulated, multi-tenant GenAI, governance isn't a constraint on the build — it's what decides whether you can ship. Design it into the first diagram.
AI & Engineering · 6 min
Hallucination is not a model problem — it's a system design problem
Ground the model in retrieved evidence, constrain its output, verify its claims, and measure everything. A layered defense against LLM hallucination.
AI & Engineering · 7 min
How We Actually Measure RAG Quality
RAG quality by vibes doesn't survive a second engineer. Decompose by stage, build the eval set from real failures, calibrate the LLM judge, and gate CI.
AI & Engineering · 7 min
Multilingual RAG, Where the Languages Are Code
Code is many languages plus an identifier dialect tokenizers shred. Use code-specialized embeddings, identifier-aware retrieval, and structure-aware chunking.
AI & Engineering · 2 min
Why knowledge graphs beat vector-only RAG
Vector-only RAG returns flat, context-poor chunks ranked by similarity. Knowledge graphs model entities and relationships to traverse why things connect.
AI & Engineering · 6 min
Choosing a Vector Database: Benchmarks from My Own Workloads
Every vector DB benchmark was run on someone else's data. Benchmark your own vector count, dimensions, and — above all — your real filtering pattern.
AI & Engineering · 8 min
POC to Production: What Breaks When RAG Meets Real Users
A RAG demo proves the happy path exists. Production is everything else — tracing, drift, evals, and learning to say 'I don't have that part.'
AI & Engineering · 7 min
The Hallucination Problem: What Reduced Ours, What Didn't
Most LLM hallucination is a retrieval failure in disguise. Fix the context first, force citations, and give the model a sanctioned way to say 'I don't know.'
AI & Engineering · 7 min
Cutting LLM Inference Cost ~40% Without Losing Quality
Inference cost optimization is a measurement problem in disguise. Fix the quality metric first, then trim context, route models, and cache the stable prefix.
AI & Engineering · 6 min
RAG vs Fine-Tuning vs Long Context: My Decision Framework
'Should we fine-tune?' is usually the wrong first question. Ask instead: is the gap knowledge or behavior, and how often does the answer change?