AI Engineer · LLM Systems · Agentic AI
I build production-grade LLM systems, agentic multi-agent pipelines & enterprise AI automation. 3+ years shipping real systems that cut operational overhead up to 90% and infrastructure costs by 70%.
About
I'm an AI Engineer based in Rawalpindi, Pakistan — specialising in LLM-powered systems, multi-agent orchestration, and enterprise automation. I turn cutting-edge research into reliable systems that run in production.
Currently at Ikonic, I build multi-tenant, RAG-backed LLM services on Anthropic Claude and AWS Bedrock — including SafetyNexus, a live real-time AI coaching platform with safety screening, citations, and role-based access control.
Across 3+ years I've shipped financial-intelligence platforms with fine-tuned models, LangGraph & CrewAI multi-agent systems, and real-time voice agents — consistently cutting operational overhead by up to 90%.
Technical Stack
01 · AI & ML
02 · Frameworks
03 · Automation
04 · Deployment
05 · Databases
06 · Data & Viz
Selected Work
A live real-time AI coaching platform for child-welfare and domestic-abuse caseworkers, built on the Safe & Together methodology. Streaming LLM guidance over SSE with safety/crisis-screening interrupts, inline citations, document parsing, multi-tenant RBAC — powered by Anthropic Claude on AWS Bedrock.
Visit live platform ↗A live, privacy-first conversational AI mentor that helps people think through workplace concerns and prepare for difficult conversations. Role-aware guidance (staff, leader, governance) grounded in evidence-based sources, with voice & text interaction and zero-retention sessions.
AI platform for banks: document analysis, risk forecasting & automated reporting. Fine-tuned DeepSeek with a BERT judge model, orchestrated via a CrewAI Researcher–Analyst–Reporter pipeline.
A LangGraph pipeline with Financial & Transcription agents that pull data from Microsoft Outlook, synthesise it, and auto-generate and dispatch structured client reports.
Replaced VAPI with a custom Twilio + OpenAI Realtime voice agent — bidirectional WebSocket audio, sub-second latency, live tool-calling mid-conversation.
End-to-end reporting pipeline: Gmail API ingestion, pandas preprocessing, a SQLite ledger with SHA-256 dedup, and a live React KPI dashboard — replacing manual QuickBooks reporting for a US client.
End-to-end n8n pipelines: lead intake, LLM-based decision routing, and automated CRM updates — removing manual operations work across the funnel.
A multimodal retrieval-augmented generation platform for business & financial experts — grounded, cited responses across text and visual sources.
An LLM agent that scrapes social platforms and answers natural-language questions by autonomously generating, validating, and running SQL against the collected data.
A 24/7 assistant with RAG over company knowledge bases, serving both voice and chat with real-time, accurate, guard-railed responses.
Upload an image and an LLM parses the text, updates a knowledge graph, and auto-builds relationships between extracted entities.
A no-code data-science platform: CSV upload, preprocessing, visualisation, and training of Linear Regression, KNN & Logistic Regression with interactive evaluation.
Career
Ikonic · Remote
Aridian Technologies · Rawalpindi
ZIYVO · Remote
IQRA University · Islamabad · CGPA 3.47 / 4.00
Learning
DeepLearning.AI · Oct 2025
Pipecat, WebRTC, real-time voice AI system design and deployment
LangGraph · 2025
Semantic, episodic & procedural memory-augmented AI agent architectures
LinkedIn Learning · May 2024
Excel, Power BI, data modeling, EDA, and dashboard optimization
Open to Opportunities
Available for AI Engineer roles, freelance builds, and collaborations. Based in Rawalpindi — fully remote-ready worldwide, open to relocation.
shaharyarshabbir348@gmail.com