Enterprise5 monthsLead with 3 client engineers

AI-Powered Enterprise Search

Find anything in your organization in seconds

A knowledge-management client

A 12,000-person organization had knowledge spread across 40+ systems - wikis, ticketing, drives, code, chat archives - with no unified search. New hires lost their first month rediscovering things that already existed. I built a semantic search platform that connectors-pull from each source, embeds the content into a vector store, applies row-level access controls per system, and answers natural-language questions with cited results. The hard problems were not retrieval quality - they were permissions, freshness, and trust.

This is a representative architecture study based on real project patterns. Specific metrics and client details have been generalized to protect confidentiality.

Results

What changed, in numbers

The metrics the engagement is measured by.

85%

Search Time

reduction in time-to-find

40+

Systems Connected

enterprise systems indexed

94%

Query Success

of searches find relevant results

15K+

Adoption

daily active users

Challenge

What was broken

Knowledge silos at scale. Asking 'have we ever solved X' could take a week and three Slack channels. Off-the-shelf enterprise search couldn't honor the per-system access controls, so it either over-shared (a compliance fire) or under-shared (useless). Permissions changed daily, content changed hourly, and people would stop trusting the tool the first time it returned a stale or unauthorized result.

Solution

The shape of the fix

A semantic search platform that indexes 40+ enterprise systems, honors per-system permissions at query time, blends lexical and vector retrieval, and returns AI-generated answers with citations - so users can verify before they trust.

Approach

How I tackled it

The concrete moves that took the project from broken to shipped.

1

Built source-specific connectors that respected each system's native ACLs at query time, not at index time

2

Used semantic chunking with embedding models tuned for the corpus, not generic web embeddings

3

Mixed lexical and vector retrieval with a learned re-ranker so exact-match queries still worked

4

Streamed near-real-time updates so a wiki edit was searchable within minutes

5

Added per-result citations so users could verify before they trusted

6

Personalized rankings based on team and recent-work signals without leaking access boundaries

Outcomes

What shipped, and what it changed

Measured results from the engagement, told as a story rather than a scoreboard.

  • Reduced average time-to-find on tracked queries by 85%

  • Indexed 40+ enterprise systems with continuous near-real-time updates

  • Reached 94% query-success rate on a held-out evaluation set

  • 15,000 daily active users within six months of internal rollout

  • Cut 'has anyone done this before' Slack threads by an estimated 60%

Stack

Technologies used

Linked entries open the technology page with related studies, playbooks, and notes.

Services

How I helped

The specific services involved in this engagement. Each links to a deeper breakdown.

Lessons

What I would tell the next team

The takeaways I carry into every similar engagement.

Permissions at query time is the only correct answer in the enterprise. Index-time ACLs go stale and become incidents

Citations are the difference between a search tool and a chatbot. Users will tolerate wrong answers if they can verify

The freshness budget matters more than the ranking algorithm. Stale results train users to leave

More patterns and playbooks live in Insights.

Have a similar challenge?

If any of this looks like the project on your desk, the conversation is the cheapest part. You can also browse other enterprise work or the full service list.

Command Palette

Search for a command to run...