RAG and AI search
RAG development and AI search over your own data
Most RAG systems fail at retrieval, not generation. The right paragraph was never indexed, the chunk split a table in half, the embedding search missed an exact product code, or the system returned a document the user was never allowed to see.
- Ingestion pipelines
- Chunking and metadata
- Hybrid search
- Reranking
Start a project
Step 1 of 5
Got it. An engineer is reading your brief.
You will hear from us within one business day, usually sooner. If it is urgent, email t@moher.ai.
Signs you need this
We build the retrieval pipeline first and measure it on its own. Clean ingestion, structure-aware chunking, hybrid keyword and vector search, a reranker, and permission filters applied at query time. Generation comes last, with citations back to the source.
- Answers that sound right and cite the wrong document
- Search that misses exact names, SKUs or case numbers
- PDFs and tables turned into unreadable chunks
- Users seeing content from files they can't open
- Stale answers because the index rebuilds by hand
- No way to tell if retrieval or the model is at fault
What you get
What is included
Ingestion pipelines
Connectors for docs, drives, wikis, tickets and databases, with parsing for PDFs, tables and scanned files, and incremental sync.
Chunking and metadata
Structure-aware chunks that respect headings, tables and code, tagged with source, date, owner and access control lists.
Hybrid search
BM25 keyword search combined with vector search in Postgres and pgvector, or a dedicated engine when scale calls for it.
Reranking
A reranking stage that reorders candidates by relevance to the actual question, measured against a labelled query set.
Cited answers
Every claim linked to the passage it came from. If retrieval finds nothing useful, the system says so instead of guessing.
Permissions-aware retrieval
Access checks applied at query time from your identity provider, so the index never leaks a document across teams.
How it runs
How we deliver RAG and AI search
- 01
Source and query audit
Inventory the content, the access rules and a set of real questions users ask, with known correct answers.
- 02
Retrieval baseline
Ingest, chunk and index, then measure recall and precision on the query set before any generation is added.
- 03
Answer layer
Generation with citations, refusal on weak retrieval, and a UI or API that fits where users already work.
- 04
Freshness and monitoring
Incremental sync, index health checks and retrieval quality tracked over time as content changes.
Industries we build this for
RAG and AI search questions
Something else? Email t@moher.ai and an engineer will answer it.
Do we need a vector database?
Often not. Postgres with pgvector handles most corpora into the tens of millions of chunks and keeps your data in one place. We move to a dedicated engine only when the numbers require it.
RAG or fine-tuning?
RAG when the answer depends on facts that change or must be cited. Fine-tuning when you need a consistent format, tone or narrow skill. Many systems use both, with retrieval doing the knowledge work.
How do you measure RAG quality?
Retrieval recall and precision on a labelled query set, then answer faithfulness to the retrieved sources. Measuring the two separately tells you which half to fix.
Can it respect our existing permissions?
Yes. We sync access control lists from your identity provider and source systems, and filter at query time so users only retrieve what they can already open.
How long does a RAG build take?
A first production system over one or two sources usually takes four to eight weeks, depending on how messy the documents are and how complex the access rules are.
Tell us what you need built
Four quick questions and an email. An engineer reads every brief and replies within one business day with next steps, and a written scope follows within a week.
- Written scope and price within a week
- Production deploy by week three
- Your repo and cloud from the first commit
- Senior engineers only, no account managers
Start a project
Step 1 of 5
Got it. An engineer is reading your brief.
You will hear from us within one business day, usually sooner. If it is urgent, email t@moher.ai.