Opportunity data was scattered
Tenders, procurement notices, and investment projects were published across separate federal and provincial sources. Users had to search each source independently.
Manifold Growth Partners built one searchable gateway to Canadian opportunities with Gyde
Canadian opportunity discovery
Users can search investment projects, tenders, and procurement notices collected from Canadian sources in one portal.
AI-assisted classification
Anthropic’s Claude models organize records by sector, province, and opportunity type, with uncertain records sent for review.
User and admin workspaces
Users can save searches and alerts, while admins manage sources, review records, and control what gets published.
About the client
About Manifold Growth Partners Inc.
Manifold Growth Partners is a quantitative investment firm that helps investors optimize their portfolios for consistent returns, independent of conventional market trends.
Manifold's Indian client base wanted to diversify beyond domestic markets, and Canada kept coming up as a target, but the information those investors needed to act on was scattered across federal, provincial, and municipal sites with no shared format.
Manifold brought in Gyde to build the discovery layer that made that diversification workable: one platform where investors could find, filter, and get alerted to Canadian opportunities that matched their sector and risk profile.
The problem statement
Finding the right Canadian opportunity took too much manual work
Large databases created more noise
Provincial registries contained thousands of projects, many of which did not match Manifold’s investment and supply objectives.
Source data was inconsistent
Dates, province names, sector labels, encodings, and record structures varied across sources, making reliable searching difficult.
Users and admins needed different controls
Investors and suppliers needed simple search and alerts. Admins needed source management, review queues, and control over publication.
How Gyde solved it
A reviewed AI workflow for finding opportunities in Canada
Defined what counts as a useful opportunity
Manifold and Gyde defined the investment projects, tenders, and procurement notices that belonged in the portal. A simplified 13-sector taxonomy gave users a clearer way to browse opportunities without navigating source-specific terminology.
Connected Canadian data sources
A separate worker collects new records, expires outdated listings, and prepares alerts nightly. Two automated sources are connected, with two more sources already specified and filtered for future integration.
Cleaned records before classification
The workflow standardizes dates, province names, sector labels, special characters, and text encodings before records reach the AI layer. Province detection uses full names and uppercase standalone codes, preventing words such as “on” from being incorrectly classified as Ontario.
Sorted and filtered for relevance
Records are sorted by date before limits are applied, preventing older notices from repeatedly filling the nightly feed. Large project registries are filtered for industrial, infrastructure, and energy opportunities above $50 million before they enter the portal.
Used different Claude models for different jobs
Anthropic assigns Claude models based on task complexity. Claude Haiku handles routine classification; thin or ambiguous records can be sent to Claude Sonnet 5 for a second opinion, and Claude Opus 5 supports newsletter drafts and answer summaries.
Added review before publication
AI suggests a record’s sector, province, opportunity type, and summary. Administrators can approve, edit, or reject these suggestions before publication. Nothing is automatically published simply because a model produced an answer.
Built separate user and admin access
Users can search opportunities, apply filters, save searches, and subscribe to relevant alerts. Administrators can manage data sources, taxonomies, review queues, user interests, and published listings.
Kept search shareable and costs tied to change
Every search and filter combination has a shareable URL that users can save or send to someone else. The nightly workflow skips unchanged records, so AI cost follows new content rather than the database's total size.
The outcome
A working pipeline, built to stay accurate as it scales
Checking a dozen-plus government and agency sites by hand
One platform, personalized alerts, role-based access for investors and admins
17,000+ lines of TypeScript across six packages
No consistent way to verify a listing's classification before acting on it
Every listing re-validated against its schema, then reviewed by a person before it publishes
174 tests, including migrations run against a database
Processing cost that would grow with the size of the whole corpus
Cost tied to new or changed content only, via the hash-and-skip step
~$60/month estimated for classification at current volumes
The pattern behind the result
What it takes to make public opportunity data useful
Classification determines discoverability
A record placed in the wrong sector can become invisible to the person looking for it. Classification quality directly affects whether users find useful opportunities.
Human review protects trust
AI organizes and summarizes records, but administrators control publication. Thin, ambiguous, or low-confidence records remain in the review queue until someone checks them.
Taxonomy changes affect the whole product
Moving from 22 sectors to 13 required updates across listings, prompts, filters, user interests, alerts, and saved searches. Retain older sector codes as tags so you don't lose previous information.
85%
Less registry noise after narrowing 2,249 projects to 332 relevant opportunities
Nightly
Current Canadian opportunities prioritized over stale listings
100%
Of published AI-classified listings pass administrator review
Explore more success stories
More enterprises running Gyde in production
How Accounti Cut Onboarding Time for Investors by 33%
Read the full story →︎Tossit deploys a skill-verification system to assess cook capability
Read the full story →︎Have a valuable workflow buried in scattered data?
Gyde can help you turn it into a focused AI system with clear data sources, review rules, model choices, and a practical production boundary.