Data & AI Engineer: I build data pipelines, APIs and LLM tooling, and ship them as open-source products people actually use.

My main project is psxdata, an open-source data platform for the Pakistan Stock Exchange. It runs end to end: a Python SDK that delivers clean, typed market data, a public REST API, a dbt warehouse on BigQuery, and an MCP server that lets AI assistants analyse the market with SQL.

• Data engineering: extraction from messy sources, append-only raw layers, dbt modelling (staging → marts, SCD Type-2 snapshots), warehouses on BigQuery and DuckDB/MotherDuck.
• AI engineering: MCP servers, LLM agents with tool use, and reusable analysis skills that turn a model into a domain assistant.
• Backend & APIs: FastAPI services with caching, rate limiting, typed errors and CI/CD to production.
• Infrastructure: Docker, Terraform, GCP (Cloud Run, Cloud Scheduler) and GitHub Actions with Workload Identity Federation.

flowchart LR
    SDK[psxdata<br/>Python SDK]
    SDK --> API[psxdata-api<br/>FastAPI REST]
    SDK --> WH[psx-warehouse<br/>dbt + BigQuery]
    SDK --> MCP[psxdata-mcp<br/>MCP server + DuckDB]
    WH --> MCP
    MCP --> AI[Claude / Cursor / VS Code]

Historical OHLCV, live quotes, index constituents, sectors, fundamentals, debt market and margin-eligible stocks for 1,000+ PSX tickers, returned as pandas DataFrames.

• Schema-tolerant parsing and multi-format date handling, so data stays consistent when upstream formats change
• Disk cache, retries with exponential backoff, rate limiting and proxy support (HTTP / SOCKS5)
• About 9,000 PyPI downloads a month

A public FastAPI service on top of the SDK. Live API

• Consistent response envelope with typed error codes and Retry-After handling
• Per-IP rate limiting and response caching
• Docker image on Docker Hub, CI/CD and auto-deploy, logs exported hourly to MotherDuck

Turns raw PSX data into tested, analysis-ready marts. dbt docs & lineage

• Append-only, hash-diffed raw layer, so a restated price is never silently overwritten
• dbt staging → marts: a Type-2 ticker snapshot (handles delistings), dimensions and five fact tables (OHLCV, index membership, sector daily, valuation daily, restatement history)
• BigQuery in production, DuckDB/MotherDuck for local dev
• Terraform-managed GCP: Cloud Run Job, Cloud Scheduler, least-privilege service account, keyless deploys via Workload Identity Federation

Gives Claude, Cursor and VS Code access to PSX data. The assistant loads data into a private in-memory DuckDB and answers questions with SQL.

• 11 tools (prices, screener, indices, sectors, fundamentals, sandboxed SQL) with row and time limits
• 6 analysis skills: stock tearsheets, value/dividend/momentum screens, Shariah screens, multi-stock comparisons, daily market wraps
• One-click installs: Claude Code plugin, Claude Desktop bundle (.mcpb), Cursor and VS Code
• A hosted, invite-only version with OAuth sign-in that also exposes the warehouse marts

Docs for the whole platform: psxdata.mintlify.app