Applied AI · Data preparation · Operational control
I use AI to turn repetitive document work into a repeatable operating process. Source information is extracted, cleaned, and reconciled before it feeds trackers, contract drafts, and status reports. I define what matters, check the outputs, and decide what needs action.
01 — Operational context
Multi-site gas work produces information across inspections, contracts, client records, and submission packs. The useful question is what each building still needs, which source supports its status, and who can move the next step forward.
RELAAM buildings
For the RELAAM takeover program, I maintained a tracker covering 101 buildings. It brought together inspection progress, documentation, contract submissions, and approval follow-up so missing items and dependencies could be reviewed in one place.
AI-assisted preparation and reusable Excel processing supported the information work around this program. Field completion, client responses, and authority decisions still had their own owners and timelines.
Document preparation
A separate preparation batch produced 109 annual maintenance contract (AMC) drafts using an AI-assisted extraction and transformation workflow. Cleaned information fed the master dataset and draft-generation process.
These are different measures: 109 describes a contract-draft batch; 101 describes the RELAAM program’s building tracker. A prepared draft is a working document awaiting the relevant review and submission steps.
02 — Working pipeline
I use Codex to assist with extraction and transformation, then carry the results into reusable Excel structures and automation. The value comes from a consistent path between the original records and the output someone needs to act on.
Gather the relevant documents, client records, internal trackers, and submission requirements. Establish the fields and output needed for the current task.
Use AI-assisted processing to extract information, structure the dataset, clean inconsistent entries, and filter or deduplicate records for review.
Cross-check the dataset against its sources and related records. Identify discrepancies, missing documents, dependencies, and items requiring clarification.
Feed the checked data into trackers, AMC drafts, and status summaries. Review the outputs and record the owners and next actions before circulation.
I review packs, trackers, and drafts against the information they were built from. An unclear value becomes a question to resolve; a missing document becomes a visible blocker. The operating record needs to show both the information available and the work still outstanding.
EDUCATIONAL DEMO All records below are synthetic. The sample illustrates review decisions; it contains no client documents or live AI processing.
Three incoming records describe two fictional buildings. One is a resubmitted copy.
S-01 / SAMPLE SERVICE BRIEF
Original source for DEMO-A.
S-02 / SAMPLE SERVICE BRIEF
The missing register needs a follow-up.
S-03 / RESUBMITTED COPY
A second copy of S-01, not another building.
The sample extraction includes an incorrect term and a duplicate. Structure makes them easier to review.
| Building | Extracted term | Source | Review state |
|---|---|---|---|
| DEMO-A | 6 months | S-01 | Check against source |
| DEMO-B | 12 months | S-02 | Asset register missing |
| DEMO-A | 12 months | S-03 | Repeated source record |
The 6-month value is a deliberately incorrect sample extraction. S-01 states 12 months.
Resolve what the sources support. Keep the outstanding document visible.
DUPLICATE / DEMO-A
S-03 repeats S-01. Retain one working record and keep both source references.
Decision pending
FIELD CHECK / DEMO-A
The extracted 6-month term conflicts with the source brief.
S-01 / SOURCE EXCERPT
Building: DEMO-A
Contract term: 12 months
Source check pending
MISSING DOCUMENT / DEMO-B
The asset register has not been supplied. Record who will request it.
Open · owner needed
Example decisions: retain one DEMO-A record with both references, correct its term to 12 months from S-01, and assign the missing DEMO-B register to Operations. The document remains outstanding.
The prepared dataset travels with its source references and outstanding actions.
This remains a working draft. The missing register stays open; submission and approval are separate steps.
03 — Daily practice
I apply the same approach as information arrives and priorities change: prepare the data, check its meaning, update the operational view, and communicate the action required.
Information in
I use AI-assisted extraction and data preparation to reduce repeated transcription and restructuring. I check relevance, completeness, and inconsistent entries before using the result in the working dataset.
Working view
DoE sheet exports, Excel lookups, and reusable templates feed the master tracker and status summaries. Reconciliation keeps new inputs aligned with existing records and makes gaps easier to find.
Information out
I use AI to assist with technical summaries, reports, and structured drafts, then review the facts, engineering meaning, and requested action. The final document needs to be usable by its intended reader.
Coordination
I pull out missing documents, blockers, ownership, and dependencies for the next follow-up. The resulting records support management visibility and the information needs of operations and the call center.
04 — Throughput & evidence
Reusable extraction, transformation, and templates let me process repeated document work as a batch. My review can then focus on source consistency, incomplete records, and decisions that need engineering or operational judgment.
The draft count records preparation output. The building count records tracker coverage. Neither figure measures approvals, AI accuracy, or a timed productivity gain.
| Work item | Repeatable processing | My review and decision |
|---|---|---|
| Source information | Extract and structure recurring fields across the input files. | Check relevance, completeness, and agreement with the original records. |
| Working dataset | Normalize entries, filter records, and identify duplicates for reconciliation. | Resolve inconsistent values and establish which source supports the working record. |
| Contract drafts | Reuse the prepared dataset and document structure across a batch. | Review content and missing items before the appropriate circulation or submission step. |
| Program status | Refresh the tracker, lookup results, and management summaries. | Confirm the status, identify blockers, and assign the next follow-up. |
The work produces a maintained source of operational information: checked datasets, a missing-document register, clearer readiness status, and summaries that connect management, operations, and the call center. Each output makes the next action easier to identify and follow through.