GovTech's Open Government Products team built Pair Search, a hybrid keyword-and-semantic AI engine over 30,000+ Hansard records since 1955, letting policy officers trace legislative precedent fast; early data show ~150 daily users and 84% of clicks landing in the top 10 results.
30,000+ records
Hansard records indexed (since 1955)
~150 users
Daily active users (soft-launch, 2024)
200 searches/day
Daily searches (soft-launch, 2024)
84 %
First-page result success rate
Details
Maturity
Pilot
Promoter
GovTech Singapore — Open Government Products, with the Parliament of Singapore
Pair Search is a hybrid AI search engine built by GovTech Singapore's Open Government Products (OGP) team, indexing more than 30,000 parliamentary Hansard records dating back to 1955 alongside Supreme Court judgments and legislation. It grew out of a 'Hack for Public Good' hackathon and soft-launched in early 2024 for use by policy officers.
Objectives
The tool aims to let policy officers 'trace legislative precedent fast' by making parliamentary debates and legal archives searchable far more effectively than existing government search interfaces, which OGP describes as 'of generally low quality for a myriad of reasons.'
Activities
Pair Search combines BM25 keyword search with e5 semantic embeddings, reranked using ColBERTv2 and served on Vespa.ai. It is used by policy officers from Singapore's Ministry of Law, Ministry of Communications and Information, Prime Minister's Office, and Ministry of Home Affairs.
Results
During the soft-launch period the tool logged around 150 daily active users and about 200 daily searches, with 84% of users who clicked a result reportedly finding what they needed within the first 10 results.
Conclusions
Pair Search is explicitly labelled a pilot, and no independent evaluation of its downstream effect on policy work has been published; the usage figures reflect self-reported/product-analytics data from GovTech rather than an external evaluation.
Implementation
Indicative cost
Low (< €50k) — No budget figures are disclosed; the tool originated from an internal hackathon and is built and run by GovTech's OGP team using existing government data and standard search infrastructure (Vespa.ai), consistent with a low-cost internal pilot.
Time to results
Short (< 1 year) — Soft-launched in early 2024 and still described as a pilot as of the source material, indicating a short implementation timeline so far with scale-up not yet confirmed.
Staffing & skills
GovTech Singapore, Open Government Products team (build and maintenance), Parliament of Singapore (Hansard data/collaboration), Policy officers from Ministry of Law, Ministry of Communications and Information, Prime Minister's Office, and Ministry of Home Affairs (end users)
Conditions for success
Access to digitized Hansard records back to 1955, Supreme Court judgments and legislation
Modern hybrid search stack (BM25 + e5 embeddings + ColBERTv2 reranking on Vespa.ai)
Origin in a hackathon ('Hack for Public Good') that seeded the initial build
In-house GovTech OGP capacity to iterate quickly during the pilot
Common failure modes
Explicitly labelled a pilot with no independent evaluation of downstream policy-outcome impact published
Usage metrics (DAUs, click-through) are self-reported by the building team rather than externally audited
Success metric (84% finding what they need in first 10 results) does not indicate what share of searches fail entirely
Where it fits
Governance type
central government digital agency (GovTech) with legislature partner
Scale
pilot, limited to policy officers in a handful of ministries/agencies
Income level
high-income (Singapore)
Commonly funded by
National / regional programmes
Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.
Do you run this practice?
Claim it —
verified implementers get a public contact pathway and can propose corrections.
Data sources
Where this practice's information was retrieved from, and when.