01 AI Systems Engineer · Multi-Agent Swarms & Reasoning

Building autonomous AI systems & resilient deep learning models.

Building production AI architectures that reason before they speak. Specializing in multi-agent swarms with Redis/Celery worker queues, causal cross-attention gating to eliminate LLM hallucinations, and real-time hydrological forecasting across the Indus Basin.

Explore Flagship Systems Experience & CGPA Get in Touch

Mizan System Adversarial Gating 07:15

Model Base: Mistral 7B + RAT
Hallucination Drop: -68% vs Base
MID Rating: 3.2 (Base: 1.1)
CDS / SRS Rating: 0.8 / 0.9

Mizan Validated

Systems & Empirical Benchmark Matrix

Production AI Architecture Telemetry · Quantitative Validations Across Multi-Agent Swarms & LLM Reasoning

Empirical validation and performance data for Mizan, Aegis, and multi-agent systems
System / Target	Domain	Evaluation Framework & Setting	Verified Performance	Baseline Standard	Validation
Mizan (Mistral 7B)	Legal LLM	Adversarial Causal Cross-Attention Feature Gating (Hallucination Drop -68%, MID, SRS)	MID 3.2 / SRS 0.9	MID 1.1 / SRS 0.5	Passed (↑)
Aegis: Nowshera D1	Hydrology	Lead Time D1: Nash-Sutcliffe KGE, CSI and Recall against observed discharge	KGE 0.89 / 0.84	KGE 0.73 / 0.82	CSI 0.70 / 0.41
Aegis: Tarbela D1	Hydrology	Lead Time D1: Reservoir inflow prediction across 2024/2025 cycles	KGE 0.75 / 0.83	KGE -0.66 / -0.57	Recall 0.69
Aegis: Mangla D1	Hydrology	Lead Time D1: Extreme event capture and hydrograph fidelity pathway	Recall 0.87 / 0.76	Recall 0.55 / 0.43	KGE 0.82
Aegis: Marala Surge	Arbitration	Dynamic Arbitration Gate (Judge Desk 𝛼 modulation across >16,000 m³/s surge event)	Recall 0.81 / 0.86	Recall 0.05 / 0.02	𝛼≈1.0 Active
Aegis: Nowshera D7	Hydrology	Extended Lead Time D7: Long-horizon flood forecasting and stability retention	KGE 0.79 / 0.65	KGE 0.20 / 0.27	Recall 0.64
Aegis: Tarbela D7	Hydrology	Extended Lead Time D7: Long-horizon stability vs ED-LSTM breakdown	Recall 0.64 / 0.64	Recall 0.06 / 0.05	Aegis KGE 0.45

Geological and Hydrological River Basin Survey Map

Flagship Case Study: Aegis AI Flood Forecasting & Disaster Routing

Real-time prediction pipeline for the Indus Basin using GEOGloWS V2 and NASA Power data. Pairwise multi-agent routing for 8km hazard zones and 25km safe shelters, combined with a bilingual conversational agent for emergency rescue operations.

02 Background · Education & Production Track Record

Education, Experience & Systems Engineering

Peer-reviewed deep learning researcher and software engineering student at HITEC University (CGPA 3.23) with independent MIT coursework and industry AI engineering internships.

Accepted Paper ACDSA 2027

Expert Arbitration Networks: Resolving Conflicting Objectives in Deep Learning Models

Int'l Conference on Advances in Computational Design & Systems Applications Accepted · 2027

Pioneered an arbitration network architecture that dynamically resolves conflicting objective loss surfaces in deep neural networks. Eliminates task interference and gradient degradation during multi-objective optimization, outperforming standard Pareto-gated baselines.

ACDSA 2027 Deep Learning Arbitration Networks Multi-Objective Optimization Loss Geometry PyTorch

Coursework & Certificate MIT Certified

MIT 15.773: Hands-On Deep Learning

Massachusetts Institute of Technology (MIT) Independent Coursework · 2026

Rigorous independent coursework covering deep neural network architectures, representation learning, transformer self-attention mechanisms, loss surface dynamics, and deployment of production PyTorch neural models.

MIT 15.773 Deep Learning Transformers PyTorch Independent 2026

University Degree CGPA 3.23 / 4.00

B.Sc. Software Engineering

HITEC University, Taxila Sep 2022 – July 2026

Rigorous engineering education covering Artificial Intelligence, Distributed Systems, Neural Network Architectures, and Database Engineering. Maintained a 3.23 CGPA while conducting independent research into multi-agent swarms and adversarial legal LLM reasoning frameworks.

Software Engineering CGPA 3.23 HITEC Taxila AI Systems Algorithms

Industry Experience AI Web Dev

AI Web Dev Intern

Nexium Jul 2025 – Aug 2025

Integrated LLM APIs into production web systems, shipping robust React.js components end-to-end. Translated high-fidelity Figma specifications into performant, responsive UI components within strict agile sprint cycles.

React.js LLM APIs Figma Agile Sprints Frontend Systems

Industry Experience Design-to-Code

UX / Web Dev Intern

Elevvo Pathways May 2025 – Jun 2025

Owned the full design-to-code pipeline, translating user research into deployed frontend architecture. Delivered accessible, functional front-end components and design artifacts specifically tailored for an ed-tech platform.

UX Research Design-to-Code Ed-Tech Accessibility Prototyping

Credential Certified

Google UX Design Certificate

Google / Coursera Aug 2024

Completed professional 7-course specialization in foundational UX research, wireframing, usability testing, and interactive prototyping in Figma. Bridging aesthetic polish with technical software rigor.

Google Certified Figma User Research Design Systems

AI & Neural Systems

Python PyTorch Transformers Hugging Face Unsloth RAG Pipelines Multi-Agent LLMs Groq API

Full-Stack & Systems

React.js Vite TypeScript FastAPI Redis Celery Docker Firebase Supabase

Design & Architecture

Figma UX Research Design Systems Wireframing Web Audio API Git / CI

One day, by the hour

Hydrological Season & Diagnostic Pipeline Telemetry · 2024/2025 Test Cycle

RAT Diagnostic 07:15 Mistral (RAT) Benchmark MID 3.2 · SRS 0.9 · CDS 0.8 14:10 Base vs RAT Matrix Verification EES 0.9 (Base: 0.2)

Basin Forecasts 08:40 Indus at Tarbela D1 Cycle Aegis KGE 0.83 vs ED -0.57 16:30 Kabul at Nowshera D1 Peak KGE 0.89 · CSI 0.70

Surge Arbitration 09:20 Jhelum at Mangla D1 Inflow Recall 0.87 vs ED 0.55 19:00 Marala Surge Gate (>16,000 m³/s) 𝛼≈1.0 · Recall 0.86 vs ED 0.02

Interactive Architecture Lab & Telemetry

Live verification of the Dynamic Judge Desk arbitration gate (𝛼-Desk) and multi-agent telemetry.

Mizan Legal LLM Feature Matrix: Adversarial Causal Cross-Attention improves MID from 1.1 to 3.2, SRS from 0.5 to 0.9, and reduces hallucinations by 68% in Mistral 7B.
Dynamic Judge Desk (𝛼-Desk Simulation): Modulates trust weight 𝛼: maintains 𝛼≈0.0 during low-flow periods for continuous hydrograph shape fidelity, and shifts to 𝛼≈1.0 during extreme surges (>16,000 m³/s).
Multisite Hydrological Validation: Aegis V13 achieves KGE 0.89 at Nowshera and Recall 0.87 at Mangla, while standard ED-LSTM fails into negative KGE (-0.66 to -0.70).

Interactive Arbitration Simulator

Trust Weight 𝛼 = 0.00

Manipulate river discharge below to observe how the Judge Desk modulates 𝛼 in real-time. In winter baseflow, 𝛼≈0.0 preserves continuous KGE shape fidelity. When flow exceeds 16,000 m³/s, 𝛼→1.0 triggers the Extreme-Event Recall Specialist without manual intervention.

Regime Winter Baseflow

Discharge 1,200 m³/s

Preset Regimes:

Observed Peak Discharge (Truth)

Aegis V13 (Judge Desk 𝛼)

ED-LSTM Baseline

Audit Telemetry Production Metrics 17:05

Model Base: Mistral 7B (RAT)
Hallucination Drop: -68% vs Base
Legal Matrix MID: 3.2 (Base 1.1)
Marala Surge Gate: >16,000 m³/s
Nowshera KGE: 0.89 / 0.84
Arbitration: Dynamic 𝛼-Desk

System Telemetry Verified

System Dossier

Aegis V13 & Mizan Verification · Production Cycle 2024–2026

Mizan: Adversarial Hallucination Reduction Matrix

Mizan Performance Matrix
Model	CDS	ECS	MID	SRS
Base Mistral	0.3	0.6	1.1	0.5
Mizan (RAT)	0.8	0.7	3.2	0.9

Aegis V13 vs ED-LSTM Baseline (D1 Lead Time)

Aegis V13 vs ED-LSTM across stations
Station	Aegis KGE	ED KGE	Aegis Recall	ED Recall
Tarbela	0.83	-0.57	0.69	0.00
Nowshera	0.89	0.73	0.78	0.61
Mangla	0.82	0.12	0.87	0.55
Marala	0.63	-0.59	0.86	0.02

Verified for Deployment HITEC SE · CGPA 3.23

Available for AI Engineering Roles & Research

Open to AI Engineer, Machine Learning Systems, Multi-Agent Architecture, and Full-Stack AI positions worldwide.

Multi-Agent Swarm Architectures (Redis + Celery)Included
Adversarial LLM Reasoning & Causal GatingIncluded
Real-Time Geospatial ML & Dynamic ArbitrationIncluded
Full-Stack Web (React, FastAPI, Docker, Figma)Included

Open Full-time roles & engineering collaborations