- Architected a scalable multi-agent conversational AI system via LangGraph and AI Foundry, for store associates scaled from 2 pilot stores to 500+ stores.
- Influenced leadership and stakeholders by reducing inference cost 20x by advocating for Nano model and preserving response quality through prompt engineering.
- Built multimodal RAG with semantic search index of 500+ documents and another index of 50K products on Azure AI Search, backed by an ETL pipeline that refined 170K products to 50K by eliminating duplication and cleansing data.
- Improved latency for 10,000+ concurrent users by implementing a pgVector-based semantic cache (evaluated against RedisVL), hitting a 10s streaming SLA (avg 6.5s, P90 8s, P95 9.5s) by semantically caching intent routing, soft guardrails, hard guardrails and responses as measured by Locust.
- Designed and reviewed PostgreSQL-based text-to-SQL agent for store-specific inventory, pricing, and product location queries.
- Provisioned infrastructure as code via Terraform for consistent, repeatable deployments across environments.
- Researched and implemented LLM-as-Judge, achieving 85% evaluation score across correctness, faithfulness, relevance, and completeness with detailed reasoning.
- Established and validated a continuous feedback loop with telemetry for observability and measured 80% user adoption among store associates.
- Addressed multi-turn conversation with topic anchor, preserving relevant context without full chat history, reducing hallucination and context window overflow.
- Architected a scalable multi-tenant, multi-agent framework on LangGraph and AWS Bedrock Agent Core with predefined templates, enabling teams to provision conversational AI agents, tools and knowledge-bases on demand via parameterized configuration and deploying onto Agent Core Runtime.
- Enforced RBAC/ABAC via JWT-based identity propagation and secured RAG retrieval with shared OpenSearch collection and siloed indices, enabling fine-grained, document-level access control and data isolation.
- Led an AI incubation program enabling AI native SDLC and collaborative teamwork culture, mentoring engineering teams to become AI-Native, experimenting and architecting 8 enterprise-scale AI/ML solutions in 12 months on Databricks and Azure AI Foundry around programmatic media ads, campaigns, recommendation engines, forecasting, segmentation, customer journey and loyalty by relying on Log Level Data (LLD) and behavioral vector embeddings to build Agentic AI solutions to fuel pre-sales and growth across different accounts. This helped acquired 4 new accounts which generated 10M+ in revenue in 12 months.
- Automated data discovery via scheduled GCP query jobs to inspect data flow through Airflow DAGs across Assortment, RDC, and Replenishment in supply chain system to uncover critical anomalies impacting 300K+ EDW records, 170K+ missed intake records, 70K+ stuck demand requests, and render on Tableau dashboard.
Bentonville, AR
Faizan Tariq
Full Stack Engineering | Scalable Architecture | Agentic AI Transformation & Strategy | Cloud-Native
13+ years architecting scalable, cloud-native platforms and agentic conversational AI systems, driving multi-million revenue impact through AI-native transformation at scale, cost optimization and enterprise-scale enablement. Skilled in business alignment, stakeholder communication and building AI-native engineering cultures.
Impact
Outcomes from recent Agentic AI and retail platform work.
- 500+ stores served by multi-agent system (from 2-store pilot)
- 20× inference cost reduction with Nano model + prompt engineering
- $10M+ revenue unlocked across 4 new accounts in 12 months
- 6,000+ US & Canada stores on Cloud Powered Checkout rollout
- 10s streaming SLA (avg 6.5s, P90 8s, P95 9.5s) for 10,000+ concurrent users
- 40ms promotions engine SLA with cloud and edge offline deployments
- 20ms tare weight microservice SLA via Kafka and MeghaCache
- 0.8s / 1.0s search P90/P95 SLAs on self-managed Elasticsearch
- 80% user adoption among store associates
- 85% LLM-as-Judge average score across correctness, faithfulness, relevance, and completeness
Experience
- Resolved data and latency issues on a next-gen automobile retail and repair platform by optimizing performance and scalability across 1TB+ of production data.
- Resolved Azure CosmosDB RUs by optimizing queries into point reads, reducing RU consumption by 60% (10,000 to 4,000 RUs), reducing latency and cost at scale.
- Remodeled high-latency search module by advocating and deploying self-managed Elasticsearch and Kibana via Helm on Kubernetes to resolve memory issues, enabling distributed pod-level replication for DR with automatic cluster-manager-node failover and meeting P90/P95 SLAs of 800ms/1000ms.
- Modernized legacy POS system with Cloud Powered Checkout (CPC), stateless microservices for Cart, Checkout and Payment based on event-driven architecture.
- Rolled out CPC to 6,000+ stores across the US and Canada market in a phased canary rollout including both cloud and edge-based offline deployments.
- Executed observability via Dynatrace and Splunk to enable monitoring, alerting, real-time telemetry and centralized logging.
- Implemented multi-store velocity controls for fraud mitigation, capping gift card purchases within a rolling 24-hour window across stores.
- Architected migration from legacy ETL to a real-time Kafka pipeline to publish store-specific tare weights to per-store topics and serving them via a MeghaCache backed microservice within 20ms SLA.
- Integrated BigQuery data with Looker to build transactional and business KPI dashboards for CPC.
- Modernized a legacy POS discounts and coupons system into a scalable, cloud-native microservices-based promotions engine, ensuring 40ms SLA and high availability via cloud and edge-based offline deployments, rolled out to 6,000+ stores across the US and Canada market.
- Automated ZZD with blue-green deployment to ensure high availability (HA) and release tags based systematized rollbacks to ensure disaster recovery (DR).
- Executed observability via Dynatrace and Splunk to enable monitoring, alerting, real-time telemetry and centralized logging.
- Setup DevOps and CI/CD pipelines for unit tests, functional tests, contract tests and automated performance tests.
- Built 4 iOS/Android native apps for a subscription-based multi-tenant enterprise suite covering digital gifting, enterprise messaging, prospecting, and live-streaming.
- Scaled across 5 tenants, each with 5-10K active users, on a shared multi-tenant architecture.
- Saved 50% dev effort by migrating 8 native apps across iOS/Android into 4 React Native apps, centralizing shared logic and bridging native modules where needed.
- Set up true CI/CD pipelines on Jenkins to reduce time to release and enabled 4 deployments/week.
- Setup test automation with 85% unit test coverage on Sonar, 80% functional test coverage on Cucumber, and 80% instrumentation test coverage with Appium.
Publications
Whitepapers on enterprise Agentic AI and data collaboration.
Whitepaper
AI strategy: Enterprise creativity enablement at scale
Platform-first architecture for generating, configuring, and governing multi-tenant agentic systems at enterprise scale.
Open PDF
Whitepaper
Data sharing consortium: Unlocking mutual value with secure collaboration
Privacy-preserving data collaboration using clean rooms, Delta Sharing, and vector embeddings for mutual value.
Open PDFEducation
BS (Computer Science)
FAST - NUCES Lahore
2009 - 2013
ICS (Computer Science)
Government College University (GCU), Lahore
2007 - 2009