AI Engineer Portfolio

A progressive series of AI engineering projects — from foundational LLM patterns to production-grade multi-agent systems. Each project builds on the previous, demonstrating increasing depth in system design, reliability engineering, and real-world deployment thinking.

Production-ready RAG and agent service built with LangChain and FastAPI.

Covers: Document ingestion, FAISS vector search, conversational memory, tool-use agents, prompt versioning, streaming responses (SSE), structured JSON output.

Tech: Python, FastAPI, LangChain, OpenAI, FAISS, Pydantic

Stateful multi-step agent built with LangGraph's StateGraph.

Covers: Explicit state machines, conditional edges, human-in-the-loop interrupt/resume, SQLite checkpointing for persistent state across sessions.

Tech: Python, FastAPI, LangGraph, Angular

Angular application that renders LangGraph agent execution graphs in real time.

Covers: Live agent state visualization, node/edge rendering, SSE-driven updates as the graph executes.

Tech: Angular, TypeScript, D3.js / Cytoscape

The most complete project in this portfolio. A production-grade, full-stack AI platform demonstrating every layer of enterprise AI engineering — from LLM orchestration to cloud deployment.

What it does: A user submits their birth profile. An 8-node LangGraph pipeline runs 5 domain agents in parallel (Vedic Astrology, Numerology, Palmistry, Tarot, Vastu), a meta-agent synthesises cross-domain consensus, hallucination is checked, an admin reviews and approves insights, and a branded PDF report is generated — with 30+ language translation support.

Key engineering highlights:

New files added (2025-05-28):

astro-intel-backend/
├── memory/
│   ├── episodic.py       ← multi-tenant correction store: log_correction(tenant_id, ...), retrieve_similar_corrections(tenant_id, ...), correction_stats(tenant_id), correction_stats_global()
│   └── persona.py        ← DEFAULT_PERSONA + build_tenant_context(query, intent, tenant_id) + format_for_prompt() + build_chandan_context() alias
├── routers/
│   └── feedback.py       ← /api/v1/feedback/* — 7 tenant-scoped endpoints (ctx.tenant_id passed to all DB functions)
└── tests/
    └── test_episodic_memory.py  ← 30 tests (16 original + 14 multi-tenant isolation), all passing

Modified files: database.py (init_episodic_tables on startup + live ALTER TABLE migration for tenant_id column), main.py (feedback router registered), routers/analysis.py (build_tenant_context(tenant_id=ctx.tenant_id) in /run; log_correction(tenant_id=ctx.tenant_id) in /approve), schemas/models.py (ApprovalRequest extended with edited_insights[]), metrics/collector.py (correction_stats_global() for dashboard)

Tech: Python 3.11, FastAPI, LangGraph, DeepSeek LLM, Angular 17, SQLite/PostgreSQL, Redis 7.2, Kafka (Confluent 7.6), Docker, AWS

AI-powered personal and team growth operating system using agentic workflows.

Covers: Goal decomposition, multi-step planning agents, progress tracking, structured output pipelines.

Tech: Python, FastAPI, LangChain/LangGraph, Angular

Automated report generation system using LLM pipelines.

Covers: Document analysis, structured report generation, multi-section synthesis, export workflows.

Tech: Python, FastAPI, OpenAI, Angular

AI-assisted resource allocation and optimization tool for engineering teams.

Covers: Skill matching, capacity analysis, LLM-driven recommendations, structured decision outputs.

Tech: Python, FastAPI, OpenAI, Angular

AI tutoring and knowledge assistant application.

Covers: Personalized Q&A, adaptive responses, knowledge retrieval, conversational AI patterns.

Tech: Python, FastAPI, OpenAI, Angular

This repository follows a trunk-based branching model with environment gates. Every merge to production goes through a human-approved promotion step — no direct push to main is allowed.

feature/*  ──PR──→  develop  ──PR──→  staging  ──promote.yml──→  main
hotfix/*   ─────────────────────────────────────────────────────→  main

# Start new work — always from develop
git checkout develop && git pull origin develop
git checkout -b feature/your-feature-name

# Work, commit, push
git commit -m "feat: description"
git push origin feature/your-feature-name

# Open PR: feature/your-feature-name → develop
# CI must pass (pytest + ng build) before merge is allowed

1. PR: feature/* → develop       CI gate (test.yml): pytest + ng build
2. PR: develop  → staging        CI gate again + auto-deploy to staging ECS
3. Verify staging manually        https://staging.aurawithrav.com
4. Run promote.yml (manual)       GitHub Actions → requires production approver
   └─ Verifies staging ECS health
   └─ Merges staging → main
   └─ Triggers build-push.yml on main
   └─ Triggers deploy.yml → prod ECS rolling update
   └─ Syncs develop with main

Always reference images by SHA tag in production — SHA tags are immutable; :latest is not.

Full details: see BRANCH_STRATEGY.md

ai-engineer/                          ← repo root (monorepo)
├── astro-intel/                      ← Angular 17 frontend (AstroIntel 360°)
├── astro-intel-backend/              ← FastAPI + LangGraph backend (AstroIntel 360°)
│   └── docker-compose.yml            ← Enterprise stack (Kafka + Redis + UIs)
├── bench-resource-optimizer/         ← Bench project
├── langchain_project/                ← Interview demo
├── senior-ai-engineer/               ← Study materials / interview prep (12 modules)
├── .github/workflows/                ← All CI/CD workflows
├── docker-compose.yml                ← Simple dev stack (SQLite, no Kafka/Redis)
├── BRANCH_STRATEGY.md                ← Full branching strategy documentation
├── PRODUCTION_DEPLOYMENT_GUIDE.md    ← AWS/ECS deployment guide
└── README.md                         ← This file

Two docker-compose files:

Phase 1 — Foundation
  langchain_project       Basic RAG + agents + streaming

Phase 2 — State & Orchestration
  langraph_project        Stateful agents, interrupt/resume
  graph-visualizer        Real-time agent graph visualization

Phase 3 — Production Multi-Agent
  astro-intel             Parallel agents, consensus, guardrails, Kafka, Redis

Phase 4 — Domain Applications
  agentic-growth-os       Growth planning automation
  ai-report-app           Document intelligence
  bench-resource-optimizer  Resource optimization
  guru-app                Adaptive tutoring

Rav Singh Chandan — Senior AI Engineer

6+ years background in Java, Spring Boot, Angular, DevOps, and Cloud (AWS/GCP). Now building production AI systems: multi-agent pipelines, LLM guardrails, semantic caching, and full-stack AI applications.

The AstroIntel 360° project is the most complete demonstration of these skills — it is not a tutorial follow-along. Every component — the 8-node LangGraph graph, the 3-tier Redis cache, the enterprise Kafka pipeline, the RBAC system, the G1–G5 guardrail stack, the CI/CD pipeline — was designed and built from scratch to solve real production problems.

Available for Senior AI Engineer, AI Platform Engineer, and Full-Stack AI Engineer roles.