SBY

Dwg no. SBY-001  /  Rev C

Director of Quality · IT Director

Sieggy Bennicoff-Yundt

I build AI workflows on quality data, and I get people to use them.

Ten years at a manufacturer of respirator test equipment: servo-driven breathing simulators that test the air packs firefighters wear. I built its quality system. Then I taught the paperwork to fill itself in, and kept a human at every signature.

Detail A, individuals chartFinal test volume

USL+0.1 L UCL CL1.665 L LCL LSL-0.1 L UNITS, IN BUILD ORDER RUN RULE TARGETS CORRECTED
Every unit in spec. The chart still said no. Illustrative data. Drag the mean in detail A

1Work

Three jobs at one company.

1 2 3

Left viewLane 1

Quality systems

Built the company's ISO 9001:2015 quality management system: 216 controlled documents across six departments, covering CAPA, NCR, internal audit, calibration, design control, and V&V. My name is on the revision history of 98% of them. Own 86% of CAPAs. Hold final sign-off authority on nonconforming product dispositions.

Front viewLane 2

AI workflows and data

Direct AI coding agents (Claude Code) across 11 internal repositories, among them a QMS portal, calibrator and SPC dashboards, an RMA pipeline, and change-control tools.

Right viewLane 3

Operations and IT

IT Director: prepare and monitor the annual IT budget, and write the API scripts, data imports, and custom reports for the MRPeasy ERP.

2Cases

Four pieces of work.

Detail A / The chart caught it first

SPC, final inspection

Every unit in spec. The chart still said no.

I built an automated SPC dashboard that feeds management KPIs. A Python and Tesseract OCR pipeline seeded it, by turning image-only final-inspection reports into structured data.

Then the run rules flagged a pattern. Our own final-inspection adjustment targets were biasing units toward over-volume. Every unit was still within its limit. I corrected the targets before the automated final test went live.

Inside the limits is not the same as on target.

Detail A, individuals chartDrag the mean

Units in spec30of 30
Above center line25of 30
Longest run, one side11above
Run rule, 8 in a rowFired

Run rule fired. Every unit is still in spec. But 11 points in a row sit above the center line, and chance alone rarely does that. The process moved off target. Find the cause before a unit fails.

Why it matters. A spec limit judges one unit at a time. The run rule judges the process. It can say the process moved while every unit still passes, so you fix the cause before a unit fails.

Spec limit line
Spec limits. What the customer accepts. Past them, a unit fails.
Control limit line
Control limits. What this process does when nothing has changed. Set from the baseline.
Center line
Center line. The target. The baseline ran on it.
Process mean line
Process mean. Where the process centers now. Drag it, or use the slider.
Run of points
Run rule. 8 points in a row on one side of the center line. Chance alone rarely makes that pattern.

Illustrative data. The real records stay with the company.

Final test instrumentBell prover

ACCEPTANCE LIMIT 1.665 L ±0.1 L PER NFPA 1850
Acceptance limit, bell prover: 1.665 L, plus or minus 0.1 L. Per NFPA 1850. The control chart tracks this measurement.

Detail B / Paperwork that fills itself in

AI workflow, engineering change control

The agent drafts. People sign.

Engineering changes used to be documented ad hoc, and steps got skipped. I designed and rolled out an AI-native engineering change process. Each engineer's Claude agent reads the change workbook, flags missing fields, and drafts the form text.

The procedure sets 3 signature gates: decision, approval, and closure verification. Our Engineering Director wrote the procedure text. I designed the AI workflow and co-approved Rev C with engineering and the CEO in June 2026.

The old Excel forms became web forms that check each entry before it is saved, run from a chat interface in each engineer's Claude session.

  1. StartEngineerChat interface in a Claude session
  2. Claude agentDrafts
    • Reads the change workbook
    • Flags missing fields
    • Drafts the form text
  3. Gate 1Decision Signed
  4. Gate 2Approval Signed
  5. Gate 3Closure verification Signed
The agent prepares. A person decides at each gate.

Section C-C / Rework, by the rate

Manufacturing quality

Rework rate: 20% to 14%.

I cut the rework rate 30% in 2026, from about 20% of units built to 14%. I mapped the causes of rework, added checks upstream of the final QA test, and changed the headform pour method with the chemical supplier.

Output grew over the same period, so I track the rate per unit built, not the count.

Units built, per 100Marked: rework

202520of 100

202614of 100

Rate per unit built. Each grid is 100 units.

Detail D / Scan the part, align it to the model

Prototype, receiving inspection

A point cloud, snapped to CAD.

I prototyped a 3D-scan receiving inspection tool. It aligns the scanned point cloud of an incoming part to its CAD model (Open3D, RANSAC for the rough fit, ICP for the fine fit) and measures what is off.

It is in the feasibility phase, with 151 tests.

  1. 1Scan the incoming part
  2. 2RANSAC: rough fit
  3. 3ICP: fine fit
  4. 4Measure what is off

Scan to CADRANSAC, then ICP

Illustrative geometry.

3Also built

Smaller tools.

  1. Supplier gate

    Audited the approved supplier list in April 2026 and flagged 18% of approved suppliers as unverified. Found the root cause with 5 Whys and added a verification gate to purchasing and supplier qualification.

  2. Local-LLM assistant

    Built a local-LLM knowledge assistant (Ollama) over the QMS and product knowledge base, so customer and design data stayed on the company network. Piloted it in May 2026 and benchmarked its retrieval.

  3. Procedure checker

    Built a Python tool, AI-coded under my direction, that checks written procedures against actual records.

  4. Route planner

    Built a route planner for the shipped calibration kits; over 90 days it projects 17% fewer trips.

  5. Customer video library

    Built a library of AI-made product videos from the manuals I wrote, recorded tests, and CAD files: unboxing, setup, troubleshooting, maintenance, and how to run tests. Early estimate: 10 to 20% fewer support calls on one product line.

  6. Gage R&R

    Ran AIAG ANOVA Gage R&R studies. The latest, in September 2026, covered 28 runs, 4 units, and 3 operators.

  7. KPI targets

    Wrote the management review procedure and won CEO approval of its 14 KPI targets (May 2026).

  8. EU conformity

    Wrote the EU Declaration of Conformity for a metabolic simulator (LVD, EMC, RoHS); it supported a sale to Germany.

  9. Excel VBA, before AI

    Built customer-facing and internal tools in Excel VBA before AI coding tools existed, including a data cleaner that parses 100,000+ line raw simulator files, removes measurement artifacts, and writes results to a database.

4General notes

How I work

Notes, unless otherwise specified:

  1. Measure the rate, not the count.

    Volume moves. A rate tells you whether the process changed.

    See section C-C, rework by the rate

  2. Keep a human at the gate.

    Agents draft. People decide, approve, and verify.

    See detail B, the three gates

  3. Check the entry before it is saved.

    A form that rejects a bad value beats a review that finds it later.

    See detail B, the web forms

  4. Keep the data at home.

    Customer and design data stayed on the company network, so the model came to the data.

    See the local-LLM assistant

5Career

Revision history

One company, three titles, newest first. A part-time side role runs alongside the current one.

Revision history, newest first
RevDateTitleDescription
Rev C 2018 to present Director of Quality (also IT Director and Marketing Manager)Respirator test equipment manufacturer
  • Report to the CEO. Informally lead a team of three.
  • Represent the company to regulators, auditors, customers, and key suppliers; led an off-site customer quality audit by a respirator manufacturer.
  • Run receiving and final inspection; performed 84% of manual final inspections.
  • Cut build lead time 21% on the main product line, from a value stream map of the assembly process.
  • Own the calibration program for both product lines.
  • Test the product's customer software releases before they go public.
  • Since 2020, turn client needs into written requirements on custom projects and verify each one.
Side role Mar 2026 to present Technical Lead, Services (part-time)ARKN, a venture group
  • Replaced a 10-step manual client go-live, spread across 4 code repositories, with one onboarding workflow.
  • Built per-client and per-operator cost tracking, so each model and data dollar maps to the client it served.
  • Built a message-audit pipeline that reads about 30,000 messages an hour, at $0.004 per 1,000 messages.
  • Python, Postgres, Next.js, Docker, built with AI coding agents (Claude Code, Codex).
Rev B 2017 to 2019 Quality ManagerSame manufacturer
  • Designed the framework for the company's first ISO 9001 quality program and wrote or oversaw all its procedures, work instructions, and forms.
  • Built the manufacturing QA program: 100% final inspection, 100% in-process inspection of critical and major characteristics, and receiving sampled to ANSI Z1.4.
  • Represented the company in a NIOSH audit of quality assurance.
  • Developed KPIs and measurement systems with top management.
Rev A 2016 to 2017 Junior Engineer (contract)Same manufacturer
  • Supported the NIOSH approval submission for an escape rebreather: design FMEA, material analysis, accelerated life testing (heat, humidity, chemical-aging models) to a 10-year storage goal, and a 50+ page failure-analysis report.
Rev 0 2011 to 2016 BS, Industrial and Systems EngineeringUniversity of Florida
  • Honors Program. Minor in Statistics.
  • First place, national facility-design competition (College Industry Council on Material Handling Education): NPV $117,805, IRR 543%.
  • Top 16 of 200+ entrants, UF Big Idea competition.

6Reference documents

Research

Best Paper, ISRP 2024 (Oxford)

"Analyzing the sensitivity of carbon dioxide measurements in ISO 16900"

Co-author. I designed the experiment, analyzed the data, and graphed the results.

Poster, ISRP 2024

Breathing resistance at altitude

Co-author.

I also write the company's technical content for sales and marketing, scientific papers included.

7Specification

Toolkit

Quality
ISO 9001:2015, CAPA and NCR, internal audit, SPC, FMEA, MSA and Gage R&R, ANSI/ASQ Z1.4 and Z1.9 sampling, calibration and measurement uncertainty (GUM, ISO/IEC 17025 traceability), engineering change control, root cause analysis (5 Whys, fishbone), Six Sigma methods (DMAIC, DOE), Lean (5S, value stream mapping).
Compliance
CE marking (LVD, EMC, RoHS), EN 61010-1, EN 61326-1, ALCOA+ data integrity, NIST SP 800-171.
AI
Claude Code, Codex, local LLMs (Ollama), LLM pipelines, retrieval benchmarks, AI workflows with human approval gates.
Data and code
Python (numpy, scipy, Open3D, OCR), SQL (Postgres, SQLite), R, Excel (VBA), TypeScript and Next.js, Git, Docker, GitHub Actions.
Hardware
NI DAQ, servo drives (Teknic ClearPath, Kollmorgen AKD), pressure transducers, mass flow controllers, bell prover, 3D scanning, engineering drawings and GD&T.
Systems
MRPeasy ERP (API scripts, imports, reports); admin of every company account (Microsoft 365, Gmail, Claude, and others).

8Title block

Contact

The fastest way to reach me is email.

LocationPanama City, Florida
TitleSieggy Bennicoff-Yundt, Director of Quality and IT Director
DrawnClaude Code
CheckedSBY
Dwg no.SBY-001
RevC
NoteDesigned and built with Claude Code, the way I build most of my tools.
Scale1:1
Sheet1 of 1
Date2026
Off the clockTennis.