ASAD
AHMAD
Architecting institutional-grade AI ecosystems and high-frequency quantitative intelligence.
AI SYSTEMS ENGINEER
ABOUT
ME
I ship AI systems and quant infrastructure that run in production — not demos, not prototypes.
LLM security gateway at sub-100ms. Quant engine with WebSocket execution and live risk controls. EdTech RAG pipeline serving real users. Every system I ship handles real load, real data, and real failure modes — engineered across AI security, quantitative finance, and full-stack infrastructure.
SKILLS
Systems built, not just concepts studied. Click to deep-dive.
CORE CAPABILITIES
- LLM guardrails with real-time token inspection & policy enforcement
- Multi-vector prompt injection detection & mitigation
- PII/PHI masking with custom regex + ML classifier pipelines
- Vendor-agnostic LLM gateway orchestration (100+ endpoints)
- AI-powered phishing & anomaly detection (DistilBERT, 98%+ accuracy)
SYSTEMS BUILT
DEPTH
Sub-50ms real-time inference latency. Production-grade middleware with zero-trust architecture.
CORE CAPABILITIES
- Alpha signal generation with XGBoost, LSTM, and ensemble stacking
- SHAP-based model explainability for regulatory-grade transparency
- Asynchronous WebSocket order execution and position management
- High-frequency portfolio risk scoring with VaR and stress testing
- Spatial ML models (Random Forest) for geo-predictive intelligence
SYSTEMS BUILT
DEPTH
Low-latency async pipelines. Institutional-grade back-testing frameworks with live market feeds.
CORE CAPABILITIES
- Distributed Monorepo architecture (Next.js + Node.js) at scale
- RAG pipelines with pgvector, semantic search & intent routing
- Asynchronous job queues with BullMQ for 100+ concurrent webhooks
- SSR / ISR rendering strategies for SEO and performance optimization
- Row-Level Security (RLS) and Zod schema validation for data integrity
SYSTEMS BUILT
DEPTH
Production-grade monorepo at scale. 85%+ LLM accuracy on Hinglish dialect parsing in real-time.
CORE CAPABILITIES
- Containerized deployment pipelines using Docker + GitHub Actions CI/CD
- Serverless function architecture on Vercel and HuggingFace Spaces
- QLoRA fine-tuning of Qwen2.5 LLMs on cost-optimized cloud infra
- Redis-based caching layers for high-throughput message queues
- Cloud cost analysis with AWS/Azure cost API integration
SYSTEMS BUILT
DEPTH
Fine-tuned LLMs at minimal GPU cost. Automated right-sizing recommendations at infrastructure level.
CORE CAPABILITIES
- Geospatial analysis across 2000+ spatial cells for 9 urban zones
- Sustainability Digital Twin modeling (Carbon, Water, Energy KPIs)
- LP optimization engines for policy simulation & resource allocation
- Graph-based infrastructure network modeling (NetworkX)
- WHO-benchmarked urban planning with scenario impact simulation
SYSTEMS BUILT
DEPTH
70% R² score in urban density prediction. Real-time policy simulation affecting multi-billion public plans.
EXPERIENCE
FOUNDER
ZARIYABuilt and scaled a multi-layered EdTech platform combining AI-assisted learning with expert-led instruction, covering structured courses, book-based knowledge systems, and personalized counseling.
Engineered a scalable content delivery and monetization system with real-time payments, serving 100+ active users with high engagement.
Launch Zariya ↗FOUNDER
JACOBDREAM LLCEstablished and operated a U.S.-based Amazon FBA wholesale business, building internal systems for product sourcing, ROI analysis, and automated profit forecasting.
Designed data-driven decision pipelines to evaluate demand, competition, and margins, enabling scalable supplier partnerships and inventory planning.
INTERN
BIG FACTIONImproved web performance and built responsive UI systems while leveraging e-commerce data insights to optimize user experience and technical infrastructure.
Visit Big Faction ↗PROJECT SHOWCASE
ZENTRIS
Secured 100+ LLM endpoints with real-time DLP middleware and streaming guardrails under <100ms latency.
TRADEVEX
Built ML-driven trading system with real-time WebSocket pipelines, SHAP-explainable signals, and automated risk controls.
INFRAVISION
Analyzed 2000+ spatial cells across Delhi NCR to forecast urban growth and infrastructure deficits with 70% R² prediction accuracy.
ZARIYA
Built cited-answer RAG engine over a classical knowledge base with expert-AI escalation when confidence drops below threshold.
REVORA
Eliminated revenue leakage for cloud kitchens by parsing Hinglish WhatsApp chats into structured orders via GPT-4o with 85%+ accuracy.
MINDGUARD
Achieved 98%+ phishing detection accuracy using DistilBERT with a hybrid heuristic validation layer for zero-day threat coverage.
OPTIOPS AI
Autonomous cloud cost optimizer using fine-tuned Qwen2.5 + RAG pipelines to generate actionable right-sizing recommendations from live infra data.
SHOPPILOT
Unified 10+ retail modules (CRM, Inventory, Analytics) into a single AI-driven commerce OS with WhatsApp storefront and zero-lag state sync.