Arham HassanContact
Queen's CS (AI)

AI systems builder.

Third-year CS student. Accreva developer intern. Elvyna founder. Four AI systems built with shipping speed and real proof.

Governance

Risky AI output gets caught

RAG

Answers cite source pages

Ledger

Every write leaves proof

INPUT

Messy problem

ROUTE

AI orchestration

OUTPUT

Working system

Architecture pattern

Deterministic rules own every write. AI decides the route. Humans confirm before commits.

100% catch rate0 false positives85% citation accuracy78% page accuracy<100ms query latencyPhase C AR/AP activeRetell voice agent builtRust codebase: 2-week onboardSAP Generative AI certified20+ quadcopter flight sessions100% catch rate0 false positives85% citation accuracy78% page accuracy<100ms query latencyPhase C AR/AP activeRetell voice agent builtRust codebase: 2-week onboardSAP Generative AI certified20+ quadcopter flight sessions

Proof wall

Evidence before adjectives.

Each card shows the problem, the system built, the proof available, and the privacy boundary.

Phase C in progressPrivate code

Accreva Business Ledger

AI can't touch a ledger unless deterministic rules stand between model output and the database. Every mutation needs an owner, a scope, and an immutable trace.

Backend core

Phase B done

AR/AP

Phase C active

Open case study
Active buildPrivate code

Elvyna

A plumber on a job site doesn't pick up. The caller hangs up. That call was $800 of work. Multiply by five missed calls a week. That's the problem Elvyna is built for.

Response target

10 sec

Code visibility

Private

Open case study
Prototype and evaluationPublic proof

Zahan Ed

A confident wrong answer in a classroom is worse than no answer. Most AI tutors have no mechanism to refuse or cite sources. They just talk.

Citation accuracy

85%

Correct refusal

100%

Open case study
Public proofPublic proof

Halluci-NOT

SAP environments have approved T-codes, vendors, and policy constraints that a generic LLM doesn't know and can't enforce. One hallucinated T-code in an AI output creates an operational incident.

Catch rate

100%

False positives

0

Open case study
Active, 2025 seasonPublic-safe proof

VEX U Robotics

Competitive robotics needs real-time autonomous decisions under sensor noise, latency, and adversarial field conditions. A system that hesitates loses the match.

Rust onboarding

2 weeks

Detection

Real-time

Open case study
Also: School Work Portfolio (public-safe coursework artifacts)

Builder OS

Tools tied to shipped systems.

Agent workflows, AI APIs, automation, backend systems, robotics, and product UI mapped to portfolio proof.

AI build loop

Cursor

Primary engineering environment for all projects.

Claude

Reasoning, code architecture, and complex problem decomposition.

Copilot

IDE code completion and review across all codebases.

Codex

Agentic implementation workflow.

Antigravity

Evaluation context.

5 proofs

AI APIs

OpenAI

Halluci-NOT governance pipeline and Elvyna workflow automation.

Anthropic APIs

Resume-backed LLM API experience and Claude build-loop reasoning.

Qwen APIs

User-reported API experience. Public project tie pending.

3 proofs

Agent systems

MCP

Accreva tool layer exposes safe ledger operations to agents.

Agent orchestration

Accreva planner/executor flow with approval gates.

2 proofs

Backend

FastAPI

Domain API for Accreva and Zahan Ed. Sole DB writer pattern.

Postgres

Accreva system of record with audit-first mutation design.

SQL

Resume-backed database skill used across data and backend work.

SQLAlchemy

Accreva-style service design and ORM-backed mutation path.

4 proofs

Frontend

React

Accreva web UI and this portfolio build.

Next.js

This portfolio build.

Tailwind

This portfolio UI system.

3 proofs

Automation

n8n

Elvyna missed-call and lead workflow orchestration.

1 proofs

Comms

Twilio

Elvyna missed-call capture and SMS delivery.

Cal.com

Elvyna booking handoff in missed-call workflow design.

2 proofs

Languages

Python

Halluci-NOT, Zahan Ed, coursework, and robotics support work.

Rust

VEX U autonomy codebase onboarding and sprint-cycle work.

Java

Queen's CS coursework and resume-backed engineering skill.

C/C++

Resume-backed systems programming skill.

JavaScript/TypeScript

Frontend work across Accreva and this portfolio.

5 proofs

ML and data

PyTorch

Resume-backed ML workflow skill.

LangChain/LlamaIndex

Resume-backed LLM workflow tooling.

sentence-transformers

Zahan Ed embedding pipeline.

FAISS

Zahan Ed vector retrieval backend.

RAG

Zahan Ed citation-grounded tutoring architecture.

Embeddings

Zahan Ed retrieval and evaluation work.

Pandas

Resume lists Pandas as learning for data workflows.

7 proofs

Robotics

OpenCV

VEX U real-time visual processing.

Sensor fusion

VEX U multi-input perception work.

2 proofs

Cloud and ops

SAP AI Core/BTP

Halluci-NOT deployment and SAP Generative AI Developer certification.

Docker

Resume-backed deployment and tooling skill.

Git

Daily engineering workflow across portfolio projects.

CI/CD concepts

Resume-backed operations foundation.

Azure

Resume lists Azure as learning.

5 proofs

Governance

Structured JSON contracts

Halluci-NOT audit-friendly downstream reporting.

Audit logging

Accreva mutations and Halluci-NOT governance logs.

Severity routing

Halluci-NOT policy triage for allowed, reviewed, and blocked outputs.

3 proofs

Development

VS Code

Daily engineering environment.

1 proofs

Local AI

Ollama

Local model evaluation and offline inference testing.

1 proofs

Voice AI

Retell

Elvyna voice agent for flexible caller intake and job-request handling.

1 proofs

Operating mode

Ships 2-3x faster.

Every project runs inside an AI-augmented build loop. The tools are verified. The output is above.

Cursor

Every project in this portfolio was built inside Cursor. Context-aware code generation across full repos.

Claude

Architecture decisions, complex decomposition, cross-file reasoning. The thinking layer.

Copilot

IDE inline completion and review across all codebases, including this portfolio build.

Working systems above show the operating pattern.

Systems timeline

A long builder line.

Hardware, robotics, AI governance, education RAG, accounting agents, and AI receptionists.

2019

Autonomous quadcopter

PID tuning, safety testing, 20+ prototype flight sessions, and documentation reused by 10+ classmates.

2023

Queen's CS begins

Computer Science with AI focus at Queen's University. Built toward AI systems from day one.

Sep 2024

Halluci-NOT

LLM governance layer for SAP GenAI Hub: policy-gated output triage, 100% catch rate on test scenarios, zero false positives.

Apr 2025

Zahan Ed

Offline RAG tutor with 85% citation accuracy, 100% correct refusal, and sub-100ms latency on CPU hardware.

Sep 2025

VEX U Robotics

AI/software R&D: Rust codebase onboarded in 2 weeks, real-time object detection, sensor fusion, cross-team autonomy work.

Jan 2026

SAP Generative AI Developer

SAP Generative AI Developer certification. Rare at third-year undergrad level. Validates enterprise AI systems design.

2026

Accreva

Developer intern work across the product cycle, currently frontend-heavy, inside a ledger system with audit logs and agent workflows.

2026

Elvyna

Owned AI receptionist business for trades: Retell voice intake, missed-call capture, lead qualification, owner text brief.

Coursework bridge

Academic work tied to shipping.

Public-safe artifacts connect AI, architecture, algorithms, systems, and documentation.

CISC 352 Artificial Intelligence

Public-safe proof route through the school-work portfolio.

CISC 365 Algorithms I

Public-safe proof route through the school-work portfolio.

CISC 322 Software Architecture

Public-safe proof route through the school-work portfolio.

CISC 360 Programming Paradigms

Public-safe proof route through the school-work portfolio.

Contact dock

Third-year CS student. Accreva developer intern. Elvyna founder. One slot open.

Targeting founders, AI teams, and startup recruiters. Public email pending approval. Profiles below.