OpsentinelOpsentinel
SpecGuardOpsMindInsight EngineAbout
Memory · Cross-Job Learning

The senior op's brain
on the screen.

Tribal knowledge, but on the screen.

Every shop has one — the foreman who knew "this customer always wants tighter tolerances than they spec" and "watch the spindle on Machine 3 in the afternoon." When that person retires, the knowledge walks out. Insight Engine writes it down for you.

Start with 3 free analyses →Schedule a 30-min walkthrough

What is this?

Insight Engine is the layer underneath SpecGuard and OpsMind that remembers everything they've ever found — every flagged RFQ, every customer who slipped, every job type that lost money — and recalls it the next time it's relevant.

When a new estimator quotes a customer they've never worked with, Insight surfaces what that customer cost you last time. When a foreman re-quotes a job that looks like one from 18 months ago, Insight names what to watch out for. It's tribal knowledge, but on the screen — and it survives staff turnover.

Built for

  • Shops with 12+ months of operational history and one or two senior people who hold the institutional memory
  • Job shops doing 30+ jobs/month across multiple customers — enough recurrence for patterns to compound
  • Quality teams who've seen the same defect or customer-issue more than once
  • Operations facing imminent retirement of the senior estimator / foreman who knows everything
  • Regulated work (aerospace / defense / medical) where prior compliance issues recur with the same customers

Probably not for you if

  • Brand-new shops with less than ~3 months of job history (no patterns to learn from yet)
  • Operations where every job is genuinely one-off and patterns can't compound
  • Teams looking for a static knowledge base — Insight Engine is dynamic, derived from analyses
  • Shops without SpecGuard or OpsMind in use — Insight is the memory layer for both, not a standalone tool

Walk through what Insight remembers

1Patterns library
opsentinel.io/insight
Demo

Insight · Patterns library · 6 active

+3 promoted this month

Acme delivery slip

9 occurrences · Last seen 2d ago

CNC-03 hot-runs

7 occurrences · Last seen 1d ago

Chemical flange margin loss

12 occurrences · Last seen 8d ago

Aluminum housing budget creep

8 occurrences · Last seen 5d ago

Material recert delay (Acme)

5 occurrences · Last seen 11d ago

Surface-finish ambiguity

14 occurrences · Last seen 3d ago

Every flagged RFQ + operational alert builds the library. 6 patterns visible.
2One pattern in detail
opsentinel.io/insight/patterns/chemical-flange-margin-loss
Demo
Medium · 12 occurrencesPromoted 14 days ago

Chemical-process flange margin loss — Acme Chem

Last 3 chemical-process flanges from this customer ran 22% over budget. Material certification rework was the biggest contributor on 2 of the 3 jobs.

Recommended next time

Confirm material cert source pre-quote. Add 12% margin for cert-related delay.

Provenance

3 SpecGuard findings · 4 OpsMind alerts · 5 manual notes

Provenance, occurrence count, recommended action — fully traceable.
3Surfacing in context
opsentinel.io/specguard/results/rfq-2026-0205
Demo

SpecGuard · New RFQ · Acme Chem

Insight · Heads upFrom 3 prior jobs

Last 3 chemical-process flanges from Acme Chem ran 22% over budget. Material certification rework was the biggest contributor on 2 of 3. Recommended: confirm material cert source pre-quote and add 12% margin.

New findings (8)

Tolerance stack-up on flange OD

Material cert source not specified

NACE-MR0175 sour-service callout missing

Rev letter on print and BOM disagree

Patterns appear inline during analyses. No separate "remember to check Insight" step.
4Pattern lifecycle
opsentinel.io/insight/patterns/lifecycle
Demo

Insight · Pattern lifecycle

Detected

23

Same finding 1–2 times. Stored, not yet surfaced.

Promoted

11

Recurred 3+ times. Surfaces in context next time.

Decaying

4

Hasn't recurred in 60d. Auto-archive in 30d.

workspace_id: yours only· No cross-workspace patterns · RLS-enforced
Auto-promote on recurrence. Auto-decay when irrelevant. Workspace-scoped.

A specific scenario

A new estimator's first month.

They get an RFQ for a chemical-process flange. The drawing looks straightforward — it's the kind of part the shop has made dozens of times.

But when SpecGuard runs the analysis, Insight surfaces a pattern-forecast banner at the top of the findings linking to prior occurrences:

Heads up — last 3 chemical-process flanges from this customer (Acme Chem) ran 22% over budget on average. Top contributors: material recert, tooling change time. Recommended: confirm material cert source pre-quote, add 12% margin for cert-related delays.

The new estimator catches it. The quote goes out with the right margin and the right cert language. The customer accepts it. Three months later the job ships on time and on margin — instead of being the fourth in a row that lost money on certs.

What you get

Patterns that confirm themselves

When the same defect signature appears across multiple jobs, customers, or machines, the pattern surfaces automatically with growing confidence.

Forward-checked against new RFQs

Every incoming SpecGuard analysis is checked against your shop's growing pattern library — "this kind of job hurt us last time".

Self-cleaning memory

Stale patterns auto-dismiss at 14 days unless they're re-confirmed. The library stays useful instead of becoming a graveyard of one-time blips.

How it actually works

Fingerprint similarity

Every SpecGuard finding and OpsMind pattern is canonicalized into a fingerprint and stored workspace-scoped (RLS at the database layer). New findings cluster against existing patterns by fingerprint similarity — so a paraphrased duplicate doesn't look like a new pattern. A vector-store upgrade for richer semantic retrieval is on the near-term roadmap.

Retrieval triggers

When a new RFQ is analyzed, when a new customer is seen, when a job type recurs — Insight runs a retrieval pass and surfaces relevant prior findings inline, not in a separate feed. No “remember to check Insight” step. The relevant context shows up at the moment you need it.

Pattern lifecycle

Patterns get more visible as they recur. A pattern with two verified occurrences promotes to verified status; five or more is treated as predictive. Underlying findings auto-dismiss after 14 days unless re-confirmed — when all of a pattern's findings have aged out, the pattern decays automatically. The library stays useful instead of becoming a graveyard of one-time blips.

Per-workspace scope

Your patterns are yours. Row-level security enforces it at the database layer. Patterns never leak across customers; the index itself is workspace-scoped at retrieval time, not just application-filtered.

Under the hood

  • Cross-module: stitches SpecGuard findings + OpsMind alerts on one canonical fingerprint
  • Outcome linking — when an alert resolves, the resolution feeds back into pattern confidence
  • Forward-checked against incoming work so the next quote starts from "this hurt us last time"
  • Stale patterns auto-dismiss at 14 days — no manual hygiene required
  • Confidence score on every pattern, growing as repetitions confirm it
  • Auditable memory — every pattern entry traces back to the specific jobs that produced it

What it doesn't do (yet)

  • Doesn't surface unprompted (yet) — patterns appear in context during SpecGuard / OpsMind analyses, not as a standalone "here's what you should know today" feed. On the post-pilot roadmap.
  • Doesn't import historical data automatically — patterns build up from analyses going forward, plus optional bulk historical RFQ import for pilots.
  • Doesn't do cross-workspace patterns (and won't — your data is yours).
  • Doesn't replace human judgement — surfaces context; the operator decides what to do with it.

See it on your own data.

Insight Engine compounds with use — the more SpecGuard analyses and OpsMind alerts your shop runs, the sharper the memory gets. A pilot here is the slowest to show value (you need 30+ days of activity for patterns to stabilise) but the highest-leverage long-term — it's the moat against the same expensive mistake hitting you twice.

Start with 3 free analyses →Email us about a pilot

Pricing starts at $5,000/mo as part of the SpecGuard + OpsMind bundle. See full pricing →

US-hosted · zero-data-retention LLM · NDA before any pilot · Security overview →