Ammar Hassan

AI Engineer | RAG & Multi-Agent Systems | Knowledge Graphs

Based in: Lahore, PakistanMost recently: AI Engineering Intern, Systems Limited

Data Science student at FAST-NUCES specializing in production-ready AI systems. Experienced in architecting autonomous multi-agent pipelines, transforming unstructured data into relational knowledge graphs, and building RAG systems with a focus on observability and evaluation.

Experience

AI Engineering Intern

Systems LimitedLahoreinternshipAug 2026 – Sep 2026

Summary

Built the FBR Tax Assistant, a retrieval-augmented QA system over Pakistan's 839-page Income Tax Ordinance; hybrid search with re-ranking retrieved the correct section for 92% of test queries against 78% for keyword search alone. Instrumented the full request path with OpenTelemetry, surfacing per-step latency, token cost, and failure rates in Phoenix and Grafana dashboards with a budget guard enforced before every model call. Built an evaluation harness over a 65-question golden set using RAGAS and DeepEval to measure faithfulness, answer relevancy, context precision and recall, plus deterministic citation verification. Quantified LLM-judge variance across repeated runs to establish a noise floor, retracting two earlier findings and preventing sub-threshold differences from being reported as improvements.

Results

Achieved 92% retrieval accuracy for test queries using hybrid search with re-ranking compared to 78% for keyword search.
Established a noise floor for LLM-judge variance to prevent reporting sub-threshold differences as improvements.

RAGHybrid SearchRe-rankingOpenTelemetryPhoenixGrafanaRAGASDeepEvalLLM-judge

AI & Software Engineering Intern

ExaVerseRemoteinternshipMay 2026 – Aug 2026

Summary

Co-built FitLoop, an AI-native gym management and member-retention PWA, on a two-person team in a one-month MVP sprint using React 19, TypeScript, Tailwind v4, and Firebase. Developed an AI plan generator as a secure Cloud Function calling Google Gemini via Genkit, producing schema-validated, versioned workout and diet plans localized to regional diets and budgets. Delivered member engagement features, a multi-channel payments module with concurrency-safe reconciliation, and multi-tenant security via custom-claim roles, Firestore rules, and automated audit tests.

Results

Delivered a full MVP sprint in one month on a two-person team.

React 19TypeScriptTailwind v4FirebaseGoogle GeminiGenkitCloud FunctionsFirestore

AI & Data Engineer (Contract)

BetafitsRemotecontractNov 2025 – Dec 2025

AI & Data Engineer, Contract-Based. Built a high-performance Python parser to extract and summarize structured signals from unstructured meeting transcripts for downstream analysis. Designed an automated LangChain and Graphiti pipeline to convert raw project information into a relational knowledge graph for entity linking and relationship mapping. Enabled advanced querying across structured entities to improve traceability, context retrieval, and internal decision-making. Partnered with a senior engineer to ship production-ready features under real delivery constraints, balancing reliability, speed, and maintainability.

PythonLangChainGraphitiKnowledge GraphsEntity LinkingRelationship Mapping

Skills

Technical

SQL

Python

LangChain

Git

LangGraph

RAG

LLM evaluation

REST APIs

LLM Workflows

Prompt-driven Application Design

Flask

GitHub

React

Exploratory Data Analysis

TypeScript

RAGAS

DeepEval

AI agents

FastAPI

Firebase

vector DBs

OpenTelemetry

Phoenix

Grafana

Pytest

Node.js

Regression Analysis

GitHub Actions

MSSQL

Neural Networks

AssemblyAI

Twilio

G-Eval

PostgreSQL

Express

Statistical Inference

Hypothesis Testing

A/B Testing

C++

Tailwind CSS

Projects

AI Customer Service Calling Agent

Developerpersonal

Built a voice-based customer support agent with real-time phone interactions, speech transcription, and LLM-generated responses. Integrated Twilio, AssemblyAI, and LangChain/LangGraph to orchestrate asynchronous call flows, session persistence, and dynamic TwiML generation in Flask.

PythonFlaskTwilioAssemblyAILangChainLangGraph

CosmicVault

Full Stack Developerpersonal

Developed a full-stack astronomy logging platform with secure authentication, role-based access control, and dashboard-oriented data views. Architected a BCNF-compliant database and built 26 RESTful APIs with complex SQL joins to support scalable data management.

ReactTypeScriptNode.jsExpressMSSQLTailwind CSS

Multi-Agent Data Pipeline Debugger

Lead Developerpersonal

Built an autonomous 5-agent LangGraph system that detects, diagnoses, and proposes fixes for data pipeline failures without human intervention. Ran statistical tests including Kolmogorov-Smirnov, Z-score, and Chi-squared to detect null spikes, volume drops, schema drift, and distribution shifts, then used an LLM to generate runnable SQL or shell fixes. Compressed a 30-90 minute manual debugging workflow to under 10 seconds using a directed state graph with parallel agent fan-out, a Streamlit monitoring dashboard, and a 10-scenario evaluation harness for detection and diagnosis accuracy.

PythonLangGraphSciPyGreat ExpectationsStreamlitPlotlyPytest

Visit the site View the code

Education

B.S. in Data Science

National University of Computer and Emerging Sciences (FAST-NUCES)Data ScienceAug 2023

Recognition

Certifications

Neural Networks and Deep Learning

DeepLearning.AI2025

Google 5-Day Gen AI Intensive

Kaggle2025

What drives the work

Motivations: AI Engineering
LLM Observability
RAG Systems
Multi-Agent Systems
Knowledge Graphs
MLOps
Data Pipelines
Autonomous Agents
Backend Development

Hire AmmarReach out about a role, a contract or a conversation.For recruiters

Ammar's twin is AI, it can make mistakes.

More people on Kanvis

Browse more people Create your own Kanvis