Voice-Driven Engineering Acts & Maintenance Reporting
for a Chemical Distributor (NDA)

We helped a leading chemical distributor replace paper-based “engineering acts” (installation / commissioning, technical maintenance, service visits) with a voice-first reporting app.

Field engineers now dictate what happened; the app transcribes domain-specific terminology (equipment models, product names, CAS numbers), auto-fills the official act, and pushes a signed PDF plus structured data into the client’s systems. Result: faster paperwork, fewer errors, quicker invoicing.
The challenge
Paper forms slowed everything down. Engineers filled them after site work, admins retyped data into ERP, and small mistakes (wrong model, misspelled chemical, missing serial) created compliance risk and delayed billing. The vocabulary is niche (formulations, trade names, CAS numbers, equipment variants), and teams often work in noisy environments with spotty connectivity. The client needed:
a simple way for admins to keep the recognized vocabulary up to date.
automatic field population from internal systems,
high transcription accuracy for chemical/product names,
a hands-free way to capture reports on site,
Outputs, compliance & integrations
The system generates a PDF (localized where needed) plus structured JSON. We push both to the DMS and back into ERP/CMMS/CRM, which immediately triggers billing. Reports link to relevant SDS/REACH references. All access is RBAC-controlled; data is encrypted at rest/in transit with full auditability.
Admin console for vocabulary & templates
Operations staff can:
override any report manually,
track audit logs and export data.
set validation rules (e.g., “CAS must be NNN-NN-N”),
manage form templates and required fields per act type,
edit keywords/synonyms and upload CSV lists of product names,
Auto-assembly of the engineering act
After transcription, an extraction layer (rules + NER) pulls entities (customer, site, equipment, lot/serial, actions performed, consumables, readings, next maintenance date). We fetch context from ERP/CMMS/CRM (work order, installed base, SLA) and pre-fill the official template:
Technical maintenance act
Corrective service act

Engineers review, add photos, and sign on the device; the customer can countersign on the same screen.
Installation/commissioning act
Domain-adapted speech-to-text (STT)
We fine-tuned the acoustic/language stack with:
recognition of CAS numbers and measurement units,
on-device noise handling and VAD (voice activity detection).
If the system is unsure, it proposes top alternatives for quick tap-to-fix.
custom vocabulary & phrase hints for product lines, equipment models, and chemicals,
Mobile & web app for engineers (personal account)
A clean “My Jobs” workspace with offline mode. Engineers can speak their notes (or type) and attach photos or barcodes. The app guides them through required fields but stays out of the way.
What we built
05
A PDF + structured data syncs to ERP/DMS → billing kicks off automatically.
04
Engineer reviews, adds photos, signs; customer signs.
03
The app auto-fills the act using transcript + ERP context.
02
STT transcribes with domain hints; doubtful terms show quick suggestions.
01
Engineer opens the job → taps Record and talks
How it works
Domain-adapted STT with custom lexicons (chemicals, CAS, product SKUs, equipment).
Observability for latency, transcription confidence, correction rates, and export success.
Template engine for acts; versioning and per-country variants.
Hybrid extraction (pattern rules + NER) with confidence scoring and user prompts.
Technology highlights
Results
The voice-first app turned field reports into same-day, error-resistant engineering acts that flow straight into the client’s systems. Engineers finish paperwork on site, even in noisy or offline conditions, while admins stop retyping and focus on exceptions. Product names, CAS numbers, and equipment details are captured correctly, templates stay current, and every change is auditable.

The result is faster approvals and invoicing, fewer compliance headaches, and a cleaner operational record across service visits.
FastAPI
React
Docker
Nginx
Redis
Kubernetes
Technology Stack:
S3
RabbitMQ
Extraction module
Entity linking
Tools calling
LLM Stack:
Project Manager
2 Backend Developers
LLM Engineer
Frontend Developer
ML Engineer
1 Frontend Develop
DevOps Engineer
Team:
Ready to map your first (or next) AI assistant?
Get a free 30-minute scoping call to pinpoint feasible use cases and a pragmatic path to production.