Ahsan Shuja

I designed and implemented a production-grade AI platform using RAG and AI agents to automate document understanding, decision support, and workflow execution. The system ingests enterprise documents, performs OCR and semantic chunking, generates embeddings, and indexes content for hybrid retrieval (BM25 and vector search). AI agents orchestrate multi-step reasoning tasks such as email classification, data extraction, and structured response generation.

I implemented evaluation pipelines and guardrails to reduce hallucinations and ensure reliable outputs. I also built scalable APIs using Python and FastAPI, integrated Azure OpenAI and Azure AI Search, and deployed using Docker with CI/CD pipelines. The platform reduced manual processing time from 15 – 20 minutes to approximately one minute and enabled consistent, auditable AI-assisted decision-making across business workflows.

Enterprise AI-powered legal contract analysis platform for the Aviation Industry, built on Azure with a retrieval-augmented generation (RAG) architecture. The platform enables legal teams to upload contracts, chat with AI answers grounded in document content with inline citations, and automate DocuSign workflows.

As a full-stack AI developer, I designed a multi-agent RAG system where a router agent classifies queries and delegates to specialized Delegation of Authority (DOA) or contract agents with strict source-only guardrails. I implemented real-time streaming chat via Server-Sent Events, a DocuSign JWT Bearer Grant integration that auto-ingests signed envelopes through webhooks, and an intelligent feedback loop where negative feedback triggers AI regeneration, admin email approval, and curated Q&A injection into future prompts. I also integrated Okta SSO alongside Azure AD, built automated SharePoint policy sync, and enforced enterprise security via Azure Key Vault, RBAC, and HMAC-verified webhooks.

Led development of an enterprise AI platform that automates receipt processing for a major airline's flight disruption claims operations (EU261/passenger compensation). The system ingests passenger-submitted receipts (meals, hotels, ground transport), uses LLM-powered extraction to parse line items, merchant data, and totals, then applies a configurable business rules engine for eligibility scoring, itemization checks, PNR validation, and per-category spend limits. A dual-pass "second opinion" verification layer cross-validates AI output to reduce hallucination risk and strengthen auditability for regulated claim decisioning. Structured outputs (categorized receipts, confidence scores, eligibility verdicts, recommended payouts) are delivered to the airline's CRM through a REST API, replacing manual officer data entry. As lead developer, I owned end-to-end architecture, prompt engineering, business rule design, CRM integration contracts, deployment pipelines across PPE/PROD environments, and direct stakeholder communication with airline leadership and operations teams during pilot validation.

Designed and architected an enterprise AI-powered log intelligence and auto-remediation platform for a Fortune 500 client's mobile back end running across two Azure Kubernetes regions. The platform processes 225,000+ daily log entries from Dynatrace, detects anomalies, groups incidents, infers root causes with confidence scoring, and generates plain-English narratives via Claude on Azure AI Foundry.

Phase 1 — Log Intelligence: CSV ingestion, Polars normalization, 7-day baselines, 6-rule anomaly detection, incident grouping, evidence-based root cause, and AI narrative. Outputs delivered via FastAPI, React dashboard, and Teams/Email/Slack alerts.

Phase 2 — AI Auto-Remediation: a Claude Code agent receives validated incidents, analyzes the codebase, implements fixes on feature branches, runs tests, and opens PRs for human approval with triage gate, scope resolver, and fix-history audit rail for governance.