Shahreyar Ashraf

Shahreyar Ashraf

Full-Stack Founder and AI Strategist specializing in scalable fintech and community platforms.

Based in: Lahore, PakistanCurrently: Product Specialist, ClarisyncExperience: 2 years

Hire me

Shahreyar is a Computer Science graduate from LUMS with a published research background in AI coding agents and extensive experience building full-stack applications. He has successfully led engineering teams and developed complex systems ranging from real-time fraud detection services to cross-platform mobile marketplaces.

Experience

Product Specialist

ClarisyncLahore, PakistanworkJun 2026 – Present

Summary

Built a shift-scheduling module for C-Sync SiteOps, gathering requirements directly from a data centre client and coordinating developers and QA through release. Own module specifications and UAT test & sign-off packages for the C-Sync Ops Hub platform, and manage configuration across multiple deployed client sites.

What I did

  • Built a CrewAI pipeline for the company's HR-tech newsletter where agents performed retrieval, filtering, and drafting.
  • Developed a set of prospect-research agents for sales to gather and structure firmographics from public sources.
  • Gathered requirements directly from the client's operations team to build a shift-scheduling module for data center critical infrastructure.
  • Turned operational requirements into clear specifications and coordinated developers and QA through the build process.
  • Owned the UAT test and sign-off package to ensure formal client acceptance before go-live.
  • Managed configuration across multiple sites as a separate part of the role.
CrewAILLM agentsUATRequirements Gathering

Founder & Full-Stack Developer

BarterEase - Strata LabsLahore, PakistanfounderJun 2025 – Present

Summary

Founded and built a full-stack platform for trading goods and services within trusted communities: a React Native (Expo) mobile app, a Node.js/Express REST API, and a Next.js admin dashboard for analytics, moderation and dispute resolution. Modelled the full exchange lifecycle (request → accept → QR-verified handoff → rating), including temporary lending with due dates and return tracking driven by scheduled cron jobs. Designed a PostgreSQL (Supabase) schema with triggers, indexes and versioned migrations, re-platforming from a legacy MongoDB store with a backfill script for cutover; Supabase Auth for identity. Delivered real-time exchange updates and support chat over Socket.IO with a Redis adapter for multi-instance scaling, plus Redis caching and rate limiting; push notifications via a Supabase Edge Function to Firebase Cloud Messaging. Containerised with Docker Compose behind Nginx and Let's Encrypt SSL; GitHub Actions pipelines detect changed services and rebuild only those, run idempotent ledgered migrations, and tag images by git SHA for one-step rollback.

What I did

  • Soft-launched the platform on Android to specific communities in Lahore
  • Instrumented the entire user journey to track engagement and transaction data from day one
  • Built the React Native app in Expo, the Next.js admin dashboard, and the public website using TypeScript.
  • Developed an admin dashboard for analytics, moderation, dispute handling, and broadcast messaging.

Results

  • Reached about 80 users and 120-odd listings across 16 communities in Lahore
  • Facilitated 24 exchanges with 20 completed
  • Maintained roughly 77 active users and close to 1,900 sessions over 90 days
  • Tracked about 17,000 actions with a DAU/MAU around 19%
  • Achieved a 60% completion rate once an exchange is requested
React NativeExpoNode.jsExpressNext.jsPostgreSQLSupabaseMongoDBSupabase AuthSocket.IORedisSupabase Edge FunctionFirebase Cloud MessagingDocker ComposeNginxLet's Encrypt SSLGitHub ActionsTypeScript

AI & Process Strategist

RozeRemit - 365 Care Group (Remittance Fintech)HybridworkNov 2025 – Jun 2026

Summary

Designed and built a real-time customer risk-scoring service in Python (FastAPI) to detect fraud in remittance transactions, backed by a PostgreSQL/Neo4j feature store with Redis caching for low-latency lookups. Combined an ensemble of XGBoost and Isolation Forest models with graph-based risk propagation across linked customers, and exposed SHAP-based explanations through the API for compliance review. Built synthetic data pipelines generating 100k+ node transaction graphs to stress-test detection on rare fraud patterns. Shipped with Docker and CI/CD for zero-downtime deployments, with Grafana monitoring and model-registry versioning.

What I did

  • Designed the initial architecture and built the real-time risk-scoring system.
  • Generated synthetic transaction graphs of 100k+ nodes with injected patterns like mule rings, structuring, and account takeovers to validate the system.
PythonFastAPIPostgreSQLNeo4jRedisXGBoostIsolation ForestSHAPDockerGrafana

Skills

Technical

SupabaseSupabase
GitGit
PythonPython
Expo
React NativeReact Native
Node.jsNode.js
ExpressExpress
JavaScriptJavaScript
PostgreSQLPostgreSQL
DockerDocker
SQLSQL
FastAPIFastAPI
Socket.IO
RedisRedis
Next.jsNext.js
Tailwind CSSTailwind CSS
FirebaseFirebase
GitHub ActionsGitHub Actions
XGBoost
PyTorchPyTorch
Neo4j
NginxNginx
Grafana
C/C++C/C++
Scikit-learn
SHAP
LLM agents (CrewAI)
FigmaFigma
Pandas
NumPy
GraphQLGraphQL
MongoDBMongoDB
Hugging Face
Google Play Console
GCPGCP
AWS
TypeScriptTypeScript
LoRA
Model Evaluation
UAT
Isolation Forest
CrewAI
DPO
LLM agents
Requirements Gathering

Projects

Real-time Log Anomaly Detection (AIOps)

Developeracademic
Built a streaming pipeline that parses high-velocity server logs into templates (Drain3) and scores them with a self-supervised Transformer; added an online learning loop with adaptive thresholds to keep false positives low.
PythonPyTorchDrain3Self-supervised Transformer

Local Coding Agent on SWE-bench Verified

personal
Ran a 4-bit quantized Qwen 3.5 9B model through Ollama on a MacBook Air using Claude Code to drive the agent loop for solving software issues.
Qwen 3.5 9BOllamaClaude CodeMCP serverDocker

LoRA-based DPO alignment pipeline

personal
Implemented a 135M-parameter language model (SmolLM) from scratch in PyTorch and preference-aligned it with DPO on the Anthropic HH dataset.
PyTorchLoRADPOSmolLMGrouped-query attentionRotary position embeddings

Education

Bachelor of Science

Lahore University of Management Sciences (LUMS)Computer ScienceJul 2026

Thesis. Studying the Footprints of AI Coding Agents in Blockchain Repositories

Societies & activities

Dean's Honour List (Spring 2025)

Recognition

Awards

Dean's Honour List

Lahore University of Management Sciences (LUMS)2025

Spring 2025

Patents & publications

Studying the Footprints of AI Coding Agents in Blockchain Repositories

publicationMSR 20262026

Mined and analysed 771 pull requests from Cursor, Copilot and Devin across 125 security-critical repositories.

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

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

More people on Kanvis

Browse more peopleCreate your own Kanvis