Cruq AI

Case studies

AI systems we have shipped

A selection of production AI systems the Cruq AI team has designed and built, across enterprise events, manufacturing, advertising, hiring, and real estate. Each is a real engagement; client names and identifying details are withheld, and the technical substance is kept intact.

Business events & membership network · Netherlands

AI matchmaking that ranks deals, not keywords

The challenge

A membership network that runs in-person business events wanted to pair attendees for high-value 1:1 conversations. Its naive approach matched people on shallow text similarity in registration forms, so two companies would be paired just for sharing an industry keyword. The brief was to shift from matching to deal-making: rank pairings by real commercial relevance, as an admin-only, GDPR-conscious batch process.

What we built

A batch pipeline of ingest, enrich, score, select, and outreach. An enrichment agent researches each company over the web (size, positioning, growth stage) with a confidence score and cited sources. A cheap deterministic and embedding pre-filter then shortlists candidates, and an LLM scores each pairing directionally across weighted commercial dimensions (revenue potential, decision authority, strategic fit, network value) with a written rationale. A deterministic selection layer enforces per-tier quotas, usage caps, and dedup so results are auditable and repeatable, not left to the model. Finally the LLM drafts ready-to-send intro emails, generated bilingually in a single pass, with every claim traceable to enrichment evidence and a human sign-off before anything is sent.

Results

Runs sort 100+ participants per event into membership tiers with per-tier match quotas and staged rollout waves. Multiple scoring models were A/B compared as the weighting evolved from keyword overlap to commercial-opportunity ranking, including a move from one-directional to bidirectional matching.

Manufacturing & supply chain · India

Turning raw ERP exports into sized inventory buffers and forecasts

The challenge

A manufacturer's make-to-stock system exported millions of inventory rows every day but left the economics empty: no buffer sizes, no throughput, no realized lead times. The result was no economically-sized replenishment buffers and no usable demand forecast. The source system could only push files out, with no direct query access.

What we built

A medallion data platform (bronze to silver to gold to forecast) that first derives the economics the ERP never populated: realized lead time reconstructed from purchase-order snapshots, and net throughput computed from the actual sales feed. It rebuilds true demand by folding three separate sales channels into one signal and removing phantom demand from warehouse-bin double-counting. A managed AI forecasting function then generates per-SKU demand forecasts, and buffer recommendations are ranked by annual throughput-at-risk and served through an AI/BI dashboard business users can query in natural language.

Results

80M+ historical rows across eight sites process end to end in about 15 minutes. Buffer recommendations were de-duplicated from roughly 68,000 down to about 9,000 real location-by-SKU policies, correcting an ~11x warehouse-bin over-count. True-demand reconciliation improved from a 0.72 to a 0.96 inventory-to-sales ratio, and the top SKU's throughput-at-risk was corrected roughly 20x once real pricing and de-duplication were applied. The whole platform runs for a few dollars a month on auto-stopping serverless compute.

Digital advertising & performance marketing · India

A chat-first command center for paid ads across Meta and Google

The challenge

Marketers running paid campaigns juggle separate ad managers to write creative, launch, optimize, and report across platforms, then stitch the performance data together by hand.

What we built

A single conversational agent that drafts ad creative, builds and optimizes campaigns, answers performance questions, and schedules organic posts across Meta and Google from one chat surface. It is built for money-safety: every spend-affecting action pauses for explicit human approval rendered as Approve or Deny in the chat, and campaigns are always created paused so nothing spends until a person activates it. Each tool is scoped to a single tenant so the model can never read or write another organization's data, and a sandbox mode demos the full loop with seeded analytics before any live ad account is connected.

Results

Delivered across five build phases: authentication and tenancy, encrypted Meta and Google OAuth connectors, real campaign creation with confirm-before-spend, hourly analytics sync with an in-chat Q&A, and organic post scheduling with auto-publish.

HR-tech & recruiting

An applicant tracking system built to be driven by AI agents

The challenge

The full hiring funnel, from posting and discovery through application, screening, and interviewing, is fragmented and manual for both recruiters and candidates. The goal was an AI-native hiring workspace where AI does the heavy lifting and AI agents are first-class users of the platform.

What we built

Resume parsing with structured extraction, AI job-description generation, and an AI candidate-matching and recommendation engine. The marquee build is AI voice interviews: a candidate opens a tokenized email link with no account, holds a spoken conversation with an AI interviewer over a streamed speech-to-text to LLM to text-to-speech pipeline, and the recruiter receives a scored evaluation and full transcript. Browser-based camera integrity checks run during the session. An Agent API with an MCP server, plus a Markdown twin of every job listing, lets external AI agents discover, read, and apply to jobs directly.

Results

AI voice interviews were run end to end in production, from tokenized invite through a scored, persisted evaluation. On the hosted speech pipeline, the candidate hears the first audio of each turn in about five seconds.

Build yours on Cruq AI

The same patterns behind this work - tenant isolation, human-in-the-loop approvals, retrieval-grounded answers, evals, and full run traces - are what Cruq AI gives you out of the box. Start in the quickstart, explore the use cases, or talk to us about an engagement.

Get started

Ready to build agents you own?

Put an agent to work for sales, support, data, or voice. Start free, or talk to us about rolling Cruq AI out across your team.

Book a call ->