Enfu AI — Enterprise AI CRM & Autonomous ERP Case Study

Role: Associate Software Engineer
Software house: Enfu Solutions
Release: Production enterprise deployment
Repository: Private client repository

THE INTELLIGENT ENTERPRISE

EnfuAI.

Intelligence in.
Real operations out.

An AI sales assistant connected to CRM and ERP. Voice, images and business questions become structured workflows with deterministic guardrails.

01 / MULTIMODAL INPUT

Customer sends a Roman Urdu voice note.

Illustrative workflow · not a live AI session

02 / PROCESSING PIPELINE

01Transcribe audio
02Route to a language model
03Check ERP pricing & mask PII
04Prepare a grounded response

03 / CONNECTED OUTPUTCRM ↔ ERP

A structured sales conversation, connected to CRM and ERP.

✓ Deterministic business guardrails

AI engineering / MY CONTRIBUTION

Resilience built into the intelligence.

Gemini Flash and Flash-Lite routing combines automatic failover, token-spend quotas and asynchronous background processing.

↳ Engineered the dual-LLM routing and queue-processing pipeline.

Built AI pipelines, guardrails and the React workspace.

01 / Contribution

My part in the system.

01
Architected an AI-native CRM and ERP ecosystem for energy manufacturing, integrating Frappe CRM, ERPNext v16, and an autonomous AI sales assistant.
02
Engineered a resilient Dual-LLM routing pipeline (Gemini Flash & Flash-Lite) featuring automated failover, token-spend quotas, and asynchronous background queue processing.
03
Implemented deterministic AI guardrails & PII sanitization, preventing LLM pricing hallucinations by verifying quotes against ERP price lists and enforcing automated PII masking before API dispatch.
04
Built a Multimodal Ingestion Pipeline capable of transcribing Roman Urdu voice notes, extracting technical specs from inverter nameplate images via OCR, and parsing payment receipts.
05
Developed 'Ask Your Data' Natural Language Querying (NLQ) engine that maps executive voice/text questions to parameterized database queries and renders interactive Recharts visualizations.
06
Constructed a high-performance Single Page Application (SPA) using React 19, Vite, and Tailwind CSS 4, delivering real-time lead management, production step-time tracking, and automated agent coaching audits.

02 / Architecture

Connected by design.

How the application connects interfaces, services, and data.

01React 19 SPA (Vite + Tailwind 4)
02Meta WhatsApp Cloud API
03Dual-LLM Engine (Gemini Flash & Flash-Lite)
04Frappe CRM & ERPNext v16 Core
05Redis Queue & MariaDB

PythonFrappe v16React 19TypeScriptGoogle Gemini APIERPNext v16Tailwind CSS 4RedisMariaDBMeta WhatsApp Cloud APIViteRecharts

03 / Product scope

Core product workflows.

Autonomous AI sales assistant integrated via Meta WhatsApp Cloud API
Resilient Dual-LLM routing pipeline (Gemini Flash & Flash-Lite) with dynamic quota failovers
Deterministic ERP price list verification guardrails & automated outbound PII sanitization
Multimodal ingestion pipeline: Roman Urdu voice notes, inverter OCR specs, and payment receipts
'Ask Your Data' natural language query engine rendering interactive Recharts analytics
Real-time production step-time tracking and automated coaching audits for sales reps

04 / Delivery

In use. Still evolving.

Deployed into active enterprise production orchestrating solar and energy manufacturing operations.
Completely eliminated pricing hallucination risks via deterministic ERP price verification guardrails.
Reduced customer lead response and quote calculation turnaround from hours to seconds.
Enabled executive voice/text data queries directly over enterprise production and financial databases.

The source code and proprietary enterprise schemas are private client property of Enfu Solutions. Architectural overviews and demonstrations use approved, sanitized data.

End of case study / 03Back to selected work ↗