Singapore · Vector retrieval · RAG

Accurate search & Q&A over your internal knowledge — proven with numbers.

QCRCE builds GPU vector retrieval and RAG systems for Singapore and Southeast Asia tech teams. We work on your own documents, tickets and knowledge bases, and report retrieval quality as recall@k and p50/p95 latency on an agreed query set.

Discuss a pilot See packages

The problem

Knowledge is scattered

Answers live across wikis, PDFs, tickets and chat. When search fails, people stop searching and start asking colleagues.

RAG ships unmeasured

Many RAG and search setups go live without a number for “did we retrieve the right passage?” — so quality debates stay subjective.

Build vs buy is unclear

Vector databases, vendors and in-house builds all look similar on slides. Teams need evidence on their own data before committing budget.

How we work

1
Measure
Agree a query set (min. 30 real questions) and success criteria with your team. Baseline your current search where one exists.
2
Build
Ingest, clean and chunk your corpus, run the embedding pipeline (model choice documented), and stand up a retrieval API or notebook demo.
3
Prove
Deliver an evaluation report — recall@k and p50/p95 latency — plus handoff docs and a walkthrough, so you decide on evidence.

Packages

Fixed scope, fixed price. Prices in SGD; invoiced via our Singapore business account.

Package A · Lead offer

RAG / Internal Knowledge Pilot

SGD 8,000 2–3 weeks

Ingest up to 50k documents (~500 MB text-like corpus)
Chunking + embedding pipeline, model choice documented
Retrieval API or notebook demo + simple eval harness
Eval report: recall@k, p50/p95 latency (min. 30 queries)
Handoff doc, 60-min walkthrough, 7-day bugfix support

50% deposit before kickoff, 50% on delivery.

Package B

Vector Retrieval SDK Paid Trial

On request 2–4 weeks

QCRCE Engine GPU vector retrieval SDK/API access for the trial
Integration guide (Python preferred)
Benchmark vs an agreed baseline (e.g. FAISS / pgvector)
GPU vs CPU comparison on your or synthetic vectors
Wrap-up memo with go / no-go recommendation

Paid trial. 50% deposit before credentials are issued.

Package C

Architecture / Decision Memo

SGD 2,500 1–2 days

Review of your docs or vendor proposals
One deep-dive call
Written risks, gaps and build-vs-buy recommendation

Prepaid.

Package D

Embedding Evaluation Scripts

SGD 1,800 3–5 business days

One-shot script: clean, chunk, embed and score your data
Reports recall@k and p50/p95 latency on a fixed query set
You keep the code

Prepaid.

Evaluation method

What we measure

recall@k: Share of queries where a correct passage appears in the top k results. Ground truth is labelled with your team.
Latency p50 / p95: Median and tail query latency, measured end-to-end on the agreed environment.
Baseline comparison: Where useful, the same queries run against your current search or an open baseline (FAISS, pgvector).

How targets are set

Targets are agreed per project, before building. As an illustration only, a pilot might target recall@10 ≥ 0.85 and p95 latency ≤ 200 ms.

These are example targets, not results from a customer project.

Data handling

Your data is used only to deliver the contracted package and is never used to train public models. Third-party APIs are disclosed in the SOW, with SG/SEA regions preferred. Corpora are deleted or returned within 30 days of project close.

About

QCRCE is a small Singapore software and technical consulting company focused on vector retrieval, RAG and embedding evaluation. It is founded and run by Mark (Ma Weijia), who works directly on every engagement.

We are software and consulting only. We do not provide financial or regulated advisory services.

Company details

Legal name	QCRCE Pte. Ltd.
UEN	202630091K
Address	68 Circular Road #02-01 Singapore 049422
Founder	Mark (Ma Weijia)

Talk to us

Tell us what your team needs to find, and where the knowledge lives. We will reply with whether a package fits.

qcrcemark@gmail.com X @WeijiaMa30462