# Full Technical Context: Nagendra Vikram Dixit

Website: https://nvdixit.co.uk/
LinkedIn: https://www.linkedin.com/in/nvdixit/
GitHub: https://github.com/nagendradixit647
Location: Rugby & Leeds, United Kingdom
Status: Active · Open to Enterprise Backend & AI Roles in the UK

---

## 1. Executive Summary
Nagendra Vikram Dixit is an Enterprise AI Agent Builder and Senior Backend Engineer with nearly 6 years of continuous enterprise software delivery at Cognizant supporting major UK retail infrastructure. He specializes in designing governed multi-agent AI pipelines (LangGraph, pgvector), resilient cloud microservices (FastAPI, Java 21, Spring Boot), and production automation engines saving 250+ annual engineering hours with zero production downtime.

---

## 2. Core Technical Competencies & Toolchains
- AI & Agentic Frameworks: LangGraph (Cyclic State Machines), LangChain, pgvector (Semantic DDL RAG), ChromaDB, Prompt Invariant Engineering, AST Guardrails (SQLGlot), Human-in-the-Loop (HITL) Review Gates.
- Backend Languages & Frameworks: Python 3.12 (FastAPI, Pydantic, SQLAlchemy, Celery), Java 21 & 17 (Spring Boot, Spring Security, Hibernate), Node.js, C++ / Qt6.
- Database & Storage Systems: PostgreSQL, pgvector, Oracle DB, ChromaDB, Redis (Semantic Caching), SQLite.
- Cloud, Infrastructure & DevOps: AWS (EC2, S3, RDS, IAM, CloudWatch, Lambda), Kubernetes (K8s, kubectl, Helm), Docker, Apache Kafka, GitLab CI/CD, GitHub Actions, Netlify Edge CDN.
- Identity, Security & Governance: Keycloak IAM (OAuth2, OIDC, RBAC, Till Role Provisioning), Zero-PII Client Architecture, AST Mutation Prevention.
- Frontend & Architecture: React 19, TypeScript 5.7+, Vanilla CSS Design Tokens, Zero-FOUC Dual Theme Architecture, WCAG 2.2 AA Accessibility.

---

## 3. Flagship AI Architecture: Governed Natural Language-to-SQL Multi-Agent Engine

### Problem & Enterprise Constraints
In a major UK retail enterprise, non-technical business analysts and support teams submit hundreds of daily queries across 400+ relational tables (Checkout, Inventory, Loyalty, Payments).
1. Context Window Limit: DDL schemas for 400+ tables exceed 180,000 tokens, causing context window exhaustion and prompt degradation.
2. Mutation Risk: LLMs hallucinate destructive DDL/DML statements (UPDATE, DELETE, DROP TABLE).
3. Compliance: Different roles require strict RBAC to prevent leaking sensitive customer or payment data.

### The 5-Stage Governed Pipeline
- Stage 01 (Intent Classification & Routing): Classifies incoming natural language prompts at the FastAPI edge gateway. General inquiries are answered directly, while analytical questions are normalized and routed into the SQL state graph.
- Stage 02 (Dynamic Schema RAG via pgvector): Overcomes the 400+ table context limit by embedding table DDLs into pgvector. Retrieves top-k (k=4) relevant table schemas dynamically with join-path inference.
- Stage 03 (Multi-Agent Generation & AST Invariants): Generates PostgreSQL syntax via LangGraph. An independent validator parses the SQL into an Abstract Syntax Tree (AST), mathematically asserting root SELECT-only execution and preventing write mutations (INSERT, UPDATE, DELETE, DROP, ALTER). Auto-injects LIMIT 100 and a 3000ms execution timeout.
- Stage 04 (Governance, RBAC & HITL Review Gates): Enforces Keycloak RBAC permissions on queried tables. Any query accessing sensitive tables triggers a state interrupt and routes an approval card to human engineers before execution.
- Stage 05 (Context Management & Asynchronous Delivery): Small result sets are delivered inline with automated markdown summaries; large payloads (>100 rows) are serialized to CSV, offloaded to cloud storage, and dispatched via asynchronous webhooks.

### Measured Impact
- Latency Reduction: Reduced query turnaround time from 4–24 hours (DBA queue) to 3.8 seconds (99.9% reduction).
- Mutation Risk: 0.0% write-mutation risk mathematically enforced by AST invariants.
- Enterprise Capacity: ~£45,000/year estimated engineering capacity saved across support and DBA teams.

---

## 4. Enterprise Automation Platforms (Cognizant UK Retail Infrastructure)

### A. Keycloak Bulk Role Provisioning Workbench
- Problem: Manual role assignments for retail POS till operators took 5 minutes per user via the Keycloak GUI.
- Solution: Engineered an asynchronous Python provisioning engine using Keycloak REST APIs, Pydantic validation, and connection pooling.
- Impact: Provisioning time reduced from 300 seconds to 600 milliseconds per user (99.8% speedup), saving 79.4 engineering hours annually.

### B. Barclays Smartpay Financial Reconciliation Pipeline
- Problem: Daily manual reconciliation of card payment terminal logs against retail ledger records consumed 30 minutes every morning.
- Solution: Automated Python data extraction runner executed in isolated GitLab CI/CD pipelines with encrypted credential injection and automated error alerting.
- Impact: Saved 167.3 engineering hours annually, eliminated manual spreadsheet toil, and ensured 100% auditable accounting reconciliation.

### C. Retail Surcharge Analytics Engine
- Problem: Manual calculation of complex payment surcharges across UK store estates caused financial reporting delays.
- Solution: Automated extraction, transformation, and distribution pipeline with automated Google Apps Script and webhook notifications.
- Impact: Reclaimed 240+ engineering hours annually with zero calculation discrepancies.

---

## 5. Open-Source Frameworks & Native Software

- LangGraph on the Job (GitHub: nagendradixit647/langgraph-on-the-job):
  Autonomous multi-agent research framework featuring cyclic state machines (Plan ➔ Ingest ➔ Retrieve ➔ Synthesize ➔ Verify ➔ Audit), persistent ChromaDB vector store, claim decomposition, and 0.70 grounding score verification.
- Flip Clock (GitHub: nagendradixit647/Flip-Clock):
  Cross-platform vintage split-flap clock with hardware-accelerated 3D mechanical flip physics, multi-display timezone synchronization, Win32 sleep suppression (kernel32.dll), WiX Toolset v4 installer, and ultra-battery-efficient Android Live Wallpaper (<0.1% battery drain).
- Personal Developer Portfolio (GitHub: nagendradixit647/personal-portfolio):
  Modern web portfolio built with React 19, TypeScript 5.7+, Vite 6, and Vanilla CSS design tokens. 100% test coverage across 20 Vitest suites (102 tests), WCAG 2.2 AA compliant, and zero cumulative layout shift.

---

## 6. Academic & Professional Credentials
- Employer: Cognizant Technology Solutions (Nov 2020 – Present, ~6 Years)
- Primary Client: Major UK Retail Enterprise Infrastructure
- Education: Bachelor of Technology (B.Tech.) in Information Technology (2016 – 2020)
- University: Dr. A.P.J. Abdul Kalam Technical University (AKTU), Uttar Pradesh, India