Arshad Ahamed

Agentic AI Engineer · Applied AI Engineer · Azure

India · Relocating to UAE / GCC · +91-9731701337 · arshad@arshadify.online

Visa: requires sponsorship · Notice: immediate.

Profile

Production grade AI engineer on Azure for enterprise operations, logistics, cold chain and telecom. I understand the customer problem and the technology at the same time: deterministic code owns money, limits and permissions, AI owns extraction, retrieval and explanation, humans approve high impact actions. Open to GCC roles across the UAE, Saudi Arabia and Qatar.

Ship production grade AI on Azure: Container Apps, UAE North residency, OpenAI and open models, Docker, CI/CD, observability.
Turn business problems into governed systems that run: deterministic code owns money and permissions, AI owns extraction and explanation.
Prove impact in numbers: billing errors 6% to 0.4%, month end close 6 days to 1, 11% lower stock loss, INR 1.2L spoilage avoided monthly.
Make AI safe to deploy: least privilege, human in the loop, prompt injection defence, audit trails, eval sets and regression gates.

How I work

1. Business problem

Owner language, money and risk first

2. Clarify

Constraints, data, who approves what

3. Prototype

Thin slice in days, real data

4. Solution

Deterministic rules, AI where it earns

5. Optimization

Evals, cost, latency, adoption

Impact

6% to 0.4%

Billing error rate

6 days to 1

Month-end close

11%

Lower stock loss

INR 1.2L

Spoilage avoided monthly

Work experience

AI ConsultantBangalore, India · Jan 2023 – Present

Soudagar Traders (family owned egg distribution and food supply business) · 10 vehicles · 20+ staff · 47 manual hours removed per month

Replaced paper ledgers with software across billing, inventory, forecasting, fleet, payroll and reporting, adopted by 20+ non technical users.
Cut invoice errors from 6% to 0.4% and month-end close from 6 days to 1 by moving rules into deterministic code.
Forecast demand on perishable stock: 11% lower stock loss, about INR 1.2 lakh spoilage avoided monthly.
Set the split I still build to: code owns money and permissions, AI owns extraction and explanation.
Removed about 47 manual hours per month across purchasing, stock and reporting, reported as operating metrics.

Core skills

Agentic systems: Multi-agent orchestration, LangGraph, LangChain, MCP, tool calling, bounded autonomy, approval gates, audit trails.
Applied AI and LLMs: RAG, vector database (Azure AI Search), embeddings, grounding, document intelligence, structured extraction, evals, forecasting.
Guardrails and governance: Deterministic rules, least privilege, authz, human in the loop, prompt injection defence, retention.
Cloud and engineering: Azure, Container Apps, Docker, CI/CD, AIOps, observability, Python, FastAPI, React, REST.

AI focus areas

RAG: Chunking, embeddings, hybrid and vector search on a vector database (Azure AI Search), reranking, citations, grounded answers 93% on a 50 case eval set.
Agentic AI: Multi-agent orchestration with LangGraph and LangChain, MCP tool servers, function calling, planning, bounded autonomy, human in the loop approvals.
Prompt engineering and fine tuning: Structured output, few shot, system prompt design, LoRA and PEFT adaptation, distillation for cost.
LLMOps and AIOps: Eval sets, golden datasets, regression gates, tracing, token cost and p95 latency budgets, drift monitoring.
Responsible AI and guardrails: Prompt injection defence, PII redaction, hallucination controls, access control, audit trails, EU AI Act and NIST aligned.

Projects

HAFIZAL GHIDHA · GCC cold chain and food compliance intelligencegidha.arshadify.online

Shipment, reefer and trade document workflows for GSO aligned food safety. A rules engine issues accept, hold or reject; AI extracts documents and explains decisions in Arabic and English.

Task success 88% · Extraction accuracy 95% · Rule accuracy 98% · Grounded answers 93% · p95 latency 34s · Cost per doc INR 5 · Human review 7% · Eval set 50 cases

Python, FastAPI, React, RAG, Docker, Azure

Cold chain compliance, end to endAzure UAE NorthDeterministic coreHuman approval

Shipment and sensors

Trade documents

AI extraction

Rules engine

Accept · Hold · Reject

Bilingual decision trace

mcpgateway · Safe AI for the telecom support deskmcpgateway.arshadify.online

Azure Container Apps, UAE North. Eight MCP tools behind eight ordered authorization checks that deny by default. Refunds never reach a payment system: they become supervisor approvals with evidence.

MCP tools 8 · Authz checks 8 · Autonomous money moved GBP 0 · Credit ceiling GBP 5.00 · Calls logged 100% · Authz precision 97% · Test coverage 95% · Eval set 80 cases

MCP, FastAPI, MongoDB, Auth0, OpenTelemetry, Azure

Telecom agent, end to endAzure Container AppsDeny by defaultNo money moved

MCP client

8 authorization stages

Guardrail pipeline

Tool gateway

Records and rules

Supervisor approval

Sovereignty and privacy by design

Region and residency planning for data, embeddings, prompts, outputs and logs.
Identity, RBAC, private connectivity, secrets, encryption and deletion by design.
Compliance language kept precise: aligned with, assessed against or certified under.

Recognition

$101K Azure founders credits awarded for a supply chain AI pitch.
Recognised as a top performer building and selling enterprise AI agents by lyzr.ai.
Available for GCC roles and freelance engagements, remote or on site.

Education

Bachelor of Engineering, Computer Science · SJB Institute of Technology, Bangalore2022