# ASADULLAH SHAFIQUE **Agentic AI Engineer | Multi-Agent Systems Architect | Enterprise AI Strategist** *Bridging AI potential with financial services reality through orchestration, compliance, and scalable adoption.* Karachi, Pakistan | +92-321-3771445 | asadullahshafique@hotmail.com [GitHub](https://github.com/asadullah48) | [LinkedIn](https://linkedin.com/in/asadullah-shafique-a00679325) | [Portfolio](https://asadullahshafique-devunity.vercel.app) | [Medium](https://medium.com/@texcotembroiderysourcinghouse) --- ## PROFESSIONAL SUMMARY Agentic AI Engineer building multi-agent systems for regulated industries, backed by 25+ years of operating a manufacturing business. Most agentic pilots in financial services die at the compliance review — not because the models are weak, but because a system that cannot show *why* it reached a conclusion cannot be signed off by a second line of defence. A control a model can argue its way past is not a control. My work makes that reviewable: typed orchestration with an auditable route, a written constitution screened deterministically **before** any model call, and evaluation that reads the execution trace rather than trusting the prose. Founder of Texcot Embroidery Sourcing House and creator of the **Textile ERP Platform**, targeting fabric mills, CMT units and garment exporters across Faisalabad, Sialkot, Gujranwala, Karachi and Lahore — so the domain intelligence is built from inside an industry rather than modelled from outside one. Completed **6 consecutive Panaversity Hackathons** (Bronze → Silver → Silver → Gold → Platinum → Agent Factory) with **85% code reusability** and **zero failed attempts**. Ships at volume: **21 production multi-agent platforms**, each with its own FastAPI gateway, test suite and public repository — **263 automated tests green** across them, all open source. --- ## AGENTIC AI MASTERY MATRIX Twelve competencies, four categories. **Every one cites a path that demonstrates it** — a status nobody can check is decoration. | Agentic Orchestration | Connectivity & Protocols | Enterprise & Scale | Reliability & Safety | |---|---|---|---| | Engineering Multi-Agent Ecosystem Design (MAS) | Engineering MCP Interoperability Layers | Engineering Domain-Specific Intelligence Systems (Finance) | Implementing Deterministic Guardrails | | Designing Graph-Based Agent Workflows | Standardizing Read-Only Agent Tool Contracts | Architecting Domain-Specific ERP Intelligence (Textile) | Designing Feedback-Driven Evaluation Loops | | Implementing Agentic Coding & SDLC Standardization | Architecting Provider-Agnostic Fallback Ladders | Implementing Cloud-Native Scale-Out | Engineering Context Integrity & Trace Auditability | | **تنسيق الوكلاء** | **الاتصال والبروتوكولات** | **المؤسسات والتوسّع** | **الموثوقية والأمان** | **Evidence, row by row** | Competency | Substrate | Status | |---|---|---| | Multi-Agent Ecosystem Design (MAS) | `backend/orchestration/` — triage agent + 4 specialists, star topology, one hop | Completed ✅ | | Graph-Based Agent Workflows | LangGraph graph; Plan-Act-Verify with no draft→approve edge | Completed ✅ | | Agentic Coding & SDLC Standardization | `CLAUDE.md` — a written operating spec governing every change | Completed ✅ | | MCP Interoperability Layers | `/mcp/server` — FastMCP over Streamable HTTP, verified against a live client | Completed ✅ | | Read-Only Agent Tool Contracts | 6 tools, read-only by default; writes require a stated reason | Completed ✅ | | Provider-Agnostic Fallback Ladders | Agents SDK → LangGraph → static answers | Completed ✅ | | Domain-Specific Intelligence (Finance) | FinAgent-Nexus — <2 ms screen, 0 model calls, 94 tests | Completed ✅ | | Domain-Specific ERP Intelligence (Textile) | CMT / Textile ERP — order lifecycle, 4 auto-billing types, party ledgers | In production ✅ | | Cloud-Native Scale-Out | `k8s/` — 11 manifests: autoscaling, network policy, service monitoring | Completed ✅ | | Deterministic Guardrails | 5 written principles; verified blocking 4/4 violations with no model reachable | Completed ✅ | | Feedback-Driven Evaluation Loops | `evals/` — deterministic trace layer + LLM judge; a case passes only if both pass | Completed ✅ | | Context Integrity & Trace Auditability | Typed shared state recording `route` and `tool_calls` | Completed ✅ | --- ## AGENT ENGINEERING METHODOLOGY A three-pillar framework — a way of building agents, not a product line. **Harness — The Environment.** The safety net around the model: tools, memory, permissions, observability. An agent that forgets what it did five steps ago is a harness problem, not a model failure. **Loop — The Feedback.** How the agent evaluates its own work and improves. Evidence over confidence: a coding agent stops when the tests pass, not when it feels done. **Graph — The Flow.** Orchestration of the full workflow: branching, approvals, retries, parallel tasks. One agent is a demo; a graph with retries and approval gates is a product. --- ## FLAGSHIP PROJECTS ### The Autonomous Agent Ecosystem — 21 Production Platforms *(Shipped)* [github.com/asadullah48](https://github.com/asadullah48?tab=repositories) · [Index](https://asadullahshafique-devunity.vercel.app/#projects) **Impact:** 21 production multi-agent platforms, each with its own FastAPI gateway, test suite and public repository. 263 automated tests green in aggregate across the 21 repositories, all open source. Not twenty-one variations on one template — each platform owns a distinct problem domain and a purpose-built agent team: **LegacyX** (COBOL and legacy-Java transpilation), **SecureBridge** (zero-trust MCP/A2A security and tool-poisoning defence), **StockAI** (autonomous inventory monitoring and demand forecasting), **GraphAI** (DAG workflow orchestration with human-in-the-loop approvals), and seventeen more spanning synthetic data, accessibility, context engineering, deterministic guardrails and multi-cloud marketplace agents. Each ships Docker and Kubernetes deployment configs and serves its own dashboard locally. **Stack:** OpenAI Agents SDK, FastAPI, Python 3.12, pytest, Docker, Kubernetes ### FinAgent-Nexus — Multi-Agent Financial Intelligence *(Shipped)* *Agentic AI Adoption for Financial Services* · [github.com/asadullah48/finagent-nexus](https://github.com/asadullah48/finagent-nexus) · [Live](https://finagent-nexus.vercel.app) **Impact:** Settles 8 of 14 compliance principles by arithmetic — a full screen in under 2 ms with zero model calls, and 94 tests that pass with no API key and no network. Three specialists (MarketAnalyst, ComplianceOfficer, WealthStrategist) on a fixed Plan-Act-Verify state machine rather than a conversation. Compliance is structural: no agent holds two of the three powers — market data, setting weights, rendering the verdict — and there is deliberately no graph edge from drafting a recommendation to approving one, so verification cannot be skipped under load or disabled by a flag. Shari'ah and regulatory principles live in a versioned constitution reviewed like code. Every run writes a hash-chained, tamper-evident audit trail. **Stack:** LangGraph, Anthropic Claude, Constitutional AI, Python 3.12, Pydantic, pytest ### AI TradeFlow — Inventory & Accounting for Wholesalers *(Shipped)* [github.com/asadullah48/ai-tradeflow](https://github.com/asadullah48/ai-tradeflow) **Impact:** 90 automated tests including full-trade-cycle API integration and agent golden-question suites with tool-citation assertions. Munshi AI ships five read-only tools behind an SDK input guardrail, with a deterministic compiled-regex constitutional screen run against the raw question before any model call. Proper FIFO *udhaar* aging, not a balance heuristic. First in the "AI for Pakistan Trade" series. **Stack:** OpenAI Agents SDK, FastAPI, SQLAlchemy, Alembic, PostgreSQL, Expo ### DevUnity Platform — This portfolio, and its own proof *(Shipped)* [github.com/asadullah48/asadullahshafique_devunity](https://github.com/asadullah48/asadullahshafique_devunity) · [Live](https://asadullahshafique-devunity.vercel.app) **Impact:** A triage orchestrator over four specialists, a real MCP server on the official SDK, and a written constitution verified blocking 4 of 4 violations with no model reachable. Every agent path degrades gracefully without an API key — the deployed site never 500s or hangs because a key is missing. Bilingual EN/AR with full RTL; AI-discoverable via llms.txt, JSON-LD and dynamic OG images. **Stack:** Next.js 15, TypeScript, FastAPI, OpenAI Agents SDK, FastMCP, Render, Vercel ### Bazaar — Unified B2B + B2C Marketplace *(Shipped)* [github.com/asadullah48/bazaar](https://github.com/asadullah48/bazaar) **Impact:** 500+ verified sellers and 10K+ products across one codebase serving both retail checkout and a wholesale RFQ engine. PKR-native, three languages. **Stack:** Next.js 15, FastAPI, Supabase, PostgreSQL, Redis, Docker ### Textile ERP Platform — Pakistan *(In build)* **Founder | 2024 – Present** · [Live demo](https://cmt-stitching-asadullah-shafiques-projects.vercel.app) **Impact:** Targets Faisalabad, Sialkot, Gujranwala, Karachi and Lahore — Pakistan's full textile heartland — as multi-tenant SaaS on Kubernetes. - **Fabric Mill Module:** roll/lot management, weaving & knitting stage tracking, yarn inventory, imported fabric handling - **CMT Core:** order lifecycle, auto-billing (4 bill types), inventory with BOM, production sessions, dispatch tracking - **Financial:** party ledgers, cash-flow tracking, accounts for multi-city hubs **Stack:** Next.js 15, FastAPI, PostgreSQL, Kubernetes, Supabase ### Agent Factory (H5) — Panaversity Hackathon 2025 *(Platinum)* **Impact:** Two-tier architecture in which a General Agent manufactures Custom Agents at scale; SKILL.md as portable, reusable intelligence units. Deployed on Kubernetes + Dapr. **Stack:** Claude Code, OpenAI Agents SDK, MCP, SKILL.md, Dapr, Kubernetes --- ## PROFESSIONAL EXPERIENCE ### Founder & CEO — CMT Stitching Operations **Texcot Embroidery Sourcing House | Karachi, Pakistan | 2020 – Present** - Engineered a multi-agent orchestration layer for financial-services research and Shari'ah screening, with separation of powers enforced structurally — no agent holds two of market data, weighting and verdict — rather than by prompt - Implemented MCP interoperability so agent capability is exposed as read-only tools on a protocol server, verified against a live client, instead of as bespoke HTTP endpoints - Architected deterministic guardrails from a written constitution, screened before any model call — verified blocking 4 of 4 violations with no model reachable, so enforcement survives a provider outage - Architected the Textile ERP Platform — order lifecycle, four auto-billing types, party ledgers and BOM inventory — to digitize Pakistan's textile value chain - Founded and operate a CMT stitching unit of 30–35 staff, running the full manufacturing lifecycle from sample fabric and design analysis through machine allocation, bulk production and inspection - Directed digital marketing across social channels, property portals and lead-generation funnels for textile and GCC real-estate clients ### Marketing Manager **JK Embroidery | Karachi, Pakistan | 2016 – 2020** - Directed marketing and production planning for embroidery manufacturing operations - Engineered staff and equipment scheduling to clear production bottlenecks, implementing process improvements that held under peak order flow - Led a team of 30–35 staff, setting priorities against available resources and work-order flow ### Country Manager / Buying Agent **Steven Berry | International | 2003 – 2005** - Coordinated international shipments end to end, guiding clients through the full procurement process - Enforced international quality standards across garment and textile categories - Conducted pre-production, in-production and final inspection of apparel, accessories and textile products --- ## HACKATHON ACHIEVEMENTS **Panaversity Hackathon Series — 6 Consecutive Wins | 85% Code Reuse | Zero Failed Attempts** - **Agent Factory (H5) — 2025** | *Platinum Tier* General Agent (Claude Code) manufactures Custom Agents (OpenAI Agents SDK). SKILL.md monetization. Digital FTE deployed on Kubernetes + Dapr. 117-slide deck. - **Cloud-Native Deployment (H4) — 2025** | *Platinum Tier* Full K8s stack: Kubernetes + Docker + Dapr + Kafka + Prometheus/Grafana/Jaeger + Helm + GitHub Actions CI/CD. - **Advanced Todo (H3) — 2025** | *Gold Tier* 149 tests passing. Recurring todos, team collaboration, AI suggestions, triple-layer Constitutional AI (7 BLOCK + 5 FLAG patterns). - **AI-Powered Todo (H2) — 2025** | *Silver Tier* 89 tests passing. Full-stack Constitutional AI todo. 70% code reuse from H1. - **Course Companion FTE (H1) — 2024** | *Silver Tier* Zero-Backend-LLM architecture. Digital FTE model. 70% code reuse from H0. - **Personal AI CTO (H0) — 2024** | *Bronze Tier* Constitutional AI foundation. Established spec-first, four-session methodology. --- ## TECHNICAL SKILLS **Languages:** TypeScript, JavaScript, Python, HTML/CSS **Frontend:** Next.js 15, React 19, Tailwind CSS v4, shadcn/ui **Backend:** FastAPI, SQLAlchemy ORM, PostgreSQL, REST APIs **AI & Agents:** OpenAI Agents SDK, Claude Code / Claude API, MCP Servers, Computer Use, LangGraph, LangChain, Constitutional AI, RAG, Streaming SSE **Cloud & DevOps:** Kubernetes, Docker, Dapr, Kafka, Helm, GitHub Actions, Vercel, Koyeb, Minikube **Observability:** Prometheus, Grafana, Jaeger, Loki **Platforms & Tools:** Git, Windows 11 / WSL2 --- ## DIGITAL MARKETING SERVICES Full-spectrum digital marketing strategy for Dubai real estate, UAE construction, and Pakistani SMEs: - Social media management (Facebook, Instagram, X/Twitter, LinkedIn) + paid advertising campaigns - Property portal listings, lead generation funnels, and monthly performance reporting - Tailored strategies for GCC/expat/Pakistani diaspora investors; AED-priced service packages --- ## EDUCATION & CERTIFICATIONS ### Completed courses **AI Fluency: Framework and foundations — Claude Academy (Anthropic’s learning platform)** Verified completion badge issued to Asadullah Shafique on September 18, 2026. [Verify completion badge](https://academy.claude.com/verify/20a54be8bbdcfa54f1ffee06f6c45082) **Introduction to Claude Cowork — Claude Academy (Anthropic’s learning platform)** Verified completion badge issued to Asadullah Shafique on September 18, 2026. [Verify completion badge](https://academy.claude.com/verify/e14c137e39659a99fb79e7afc1394c09) **Claude Code 101 — Claude Academy (Anthropic’s learning platform)** Verified completion badge issued to Asadullah Shafique on September 18, 2026. [Verify completion badge](https://academy.claude.com/verify/64b43e3f91ee11516e562d2c35801e3a) **AI Fluency for builders — Claude Academy (Anthropic’s learning platform)** Verified completion badge issued to Asadullah Shafique on September 18, 2026. [Verify completion badge](https://academy.claude.com/verify/2cfb8dd802a841177aba39028ef6cf4e) **AI capabilities and limitations — Claude Academy (Anthropic’s learning platform)** Verified completion badge issued to Asadullah Shafique on September 18, 2026. [Verify completion badge](https://academy.claude.com/verify/e5f7a0e1a35e7d3e6626241402d556be) **Claude 101 — Claude Academy (Anthropic’s learning platform)** Verified completion badge issued to Asadullah Shafique on September 18, 2026. [Verify completion badge](https://academy.claude.com/verify/6d808a4f90c00a98b326959280f5c280) **Claude Code in action — Claude Academy (Anthropic’s learning platform)** [Verify completion badge](https://academy.claude.com/verify/f8192381a00c3d85b011b7cd94673286) **Introduction to Model Context Protocol — Claude Academy (Anthropic’s learning platform)** [Verify completion badge](https://academy.claude.com/verify/881db997d7957558e2d108c9ab9d174e) ### Governor Sindh Initiative for Artificial Intelligence & Computing (Panaversity) **Feb 2024 – Present** · Advanced study in Agentic AI and Generative AI **Forward Deployed Engineer Training — Track B (Accelerated), 13 weeks.** Governed Vertical Systems of Record, stateless MCP (rev. 2026-07-28), multi-agent research systems, and structured extraction pipelines with validation and human review. Coursework across TypeScript/JavaScript, React and Next.js, Python, agent SDKs and AI frameworks, and Loop / Harness / Graph-based agent architectures. **Certifications — expected, not yet awarded.** Listed by awarding body, because PCAR-F is Panaversity’s own exam aligned to Anthropic’s CCAR-F blueprint rather than an Anthropic exam: - **Sindh Board / University of Karachi** — programme certification - **Panaversity** — PCAR-F, Architect Foundations, aligned to Anthropic’s CCAR-F blueprint - **Anthropic** — [Claude Certified Associate – Foundations](https://anthropic-partners.skilljar.com/claude-certified-associate-foundations-certification) → Claude Certified Architect – Foundations ### Prior qualifications - **Alim — 5-Year Islamic Studies Course** | Burooj Institute, Karachi, Pakistan - **Associate Degree — Textile Technology** | Textile Institute of Pakistan (APTMA Project), Karachi, 1997 - **H.S.C Pre-Engineering** | Pakistan Shipowner's Government College, Karachi, 1994 - **S.S.C Science** | Ladybird Grammar School, Karachi, 1992 --- ## ONLINE PRESENCE Portfolio: [asadullahshafique-devunity.vercel.app](https://asadullahshafique-devunity.vercel.app) GitHub: [github.com/asadullah48](https://github.com/asadullah48) | Medium: [@texcotembroiderysourcinghouse](https://medium.com/@texcotembroiderysourcinghouse) X/Twitter: [@texcotembroide1](https://x.com/texcotembroide1) | Instagram: [@shafiqueasadullah](https://instagram.com/shafiqueasadullah)