Agentic AI · LangGraph · Hybrid RAG · Python Backend

Building production agentic systems with architectural rigor.

I’m Jeevan Bandanadham, an AI Engineer building production Generative AI and agentic systems with LangChain, LangGraph, and Python backends. Hands-on with hybrid RAG, LangSmith observability, cyclic state machines, and deterministic safety gates.

Explore Systems & OS View Experience Contact→

AI Engineer at Trivexa Technologies · Open for Collaborations•Andhra Pradesh, India

Jeevan Bandanadham

AI Engineer @ Trivexa Technologies

App Store & LangSmith

LangGraph StateGraphs

ChromaDB + BM25 RRF

Selected Work

Proof over claims.

A curated selection of AI systems, multi-agent state machines, and resilient backend services built with strict evaluation and measurable outcomes.

Agentic AI · Multi-Agent Orchestration•Open Source OS

№ 01

Project Anveshan

Open Deep Research Operating System

Problem: Single-prompt AI search and naive RAG suffer from severe context degradation, citation hallucination, and lack of methodical critique when investigating complex academic questions.

Engineering: Engineered an autonomous multi-agent state machine (Orchestrator, Academic Search Manager, Evidence Extractor, Adversarial Critic, and Synthesizer) with local-first Ollama execution, React + Vite SSE live streaming dashboard, and Python SDK with hierarchical LangSmith tracing.

Agent Topology5 Coordinated Agents

Citation Integrity100% Grounded (No Hallucinations)

LangGraphMulti-AgentLangSmithPython SDK+4

Examine Architecture & Case Study→

Legal NLP · Deterministic Guardrails•Educational Agent

№ 02

NyayaPath

Dual-Regime Legal Guidance Agent

Problem: India's transition from colonial statutes (IPC, CrPC) to modern criminal codes (BNS, BNSS) creates severe legal misdirection when generic LLMs confuse identical section numbers with completely different legal meanings (e.g., Section 482).

Engineering: Architected the 12-node cyclic StateGraph in Python 3.11 with FastAPI backend, ChromaDB vector store, BM25 lexical retriever, Reciprocal Rank Fusion, deterministic section allowlist, React 19 Vite TypeScript frontend, and end-to-end LangSmith telemetry.

Statutory Corpus2,820 Partitioned Sections

Graph Complexity12-Node Cyclic StateGraph

LangGraph (12-Node)ChromaDBBM25 / RRFFastAPI+4

Examine Architecture & Case Study→

Production AI · Mobile & Backend•Production · App Store

№ 03

WorldTrendz

Production Conversational AI iOS Application

Problem: E-commerce mobile shoppers face friction navigating large catalogs and require instant, context-aware conversational shopping assistance with low latency on mobile devices.

Engineering: Engineered the production AI backend microservices in Python and FastAPI, handling live user sessions, stateless JWT authentication, context window pruning, product catalog retrieval, and multi-model failover routing.

DeploymentApple App Store (Live)

Backend LatencySub-second Streaming

FastAPIProduction GenAIApple App StoreMulti-Model LLMs+3

Examine Architecture & Case Study→

Enterprise AI · Systems & Cloud•Production Service

№ 04

Production GenAI & RAG Services

Enterprise Backend Architecture at Vaaryan Interfaces

Problem: Enterprise clients require custom LLM integrations, grounded RAG search over internal knowledge bases, and strictly authenticated APIs capable of cloud scale.

Engineering: Delivered authenticated REST APIs, containerized Docker microservices deployed on Google Cloud Platform and Firebase, with persistent data storage in PostgreSQL and MongoDB, verified via PyTest.

Cloud PlatformsGCP & Firebase

DatabasesPostgreSQL & MongoDB

PythonFastAPILangChainRAG Pipelines+5

Examine Architecture & Case Study→

Detailed test suites, trace logs, and architecture documentation available upon request.View GitHub Profile →

Career Journey

Professional experience.

Engineering production AI backends, multi-model LLM orchestration, and enterprise RAG systems across commercial client applications.

AI EngineerCurrent · Full-time

Trivexa Technologiestrivexatechnologies.in

Jan 2026 – Present

Design production agentic backends with Python, LangChain, and LangGraph — multi-agent workflows, RAG pipelines, and tool-using agents.
Build FastAPI services with guardrails and tracing for long-horizon agent orchestration used in Trivexa’s AI systems.

Freelance AI & Backend EngineerProduction App Store

WorldTrendz LLCApple App Store

Mar 2026 – Jun 2026

Designed and deployed WorldTrendz, a production conversational AI app published on the Apple App Store.
Built secure FastAPI backends, authentication, context management, and multi-model LLM orchestration for live users.

AI Engineer / Associate Backend EngineerProduction Cloud

Vaaryan Interfaces Pvt. Ltd.

2025

Built production GenAI backend services with Python, FastAPI, and LangChain — RAG pipelines, embeddings, vector retrieval, and prompt engineering.
Delivered authenticated REST APIs, Dockerized services on GCP / Firebase, and maintained quality with PyTest, debugging, and code reviews.

AI Engineering InternPromotion upon Ownership

Vaaryan Interfaces Pvt. Ltd.

2024 – 2025

One-year internship in Python backend, LLM integration, and RAG; built FastAPI APIs with PostgreSQL/MongoDB; promoted based on ownership.

Verified Credentials

Certifications & specializations.

Rigorous accreditations in LangGraph state machines, deep agent architectures, LangSmith observability, and multi-tool evaluations.

LangChain Academy

Introduction to LangGraph (Python)

State, reducers, checkpointing, HITL, prompt caching

LangChain CertifiedVerify

LangChain Academy

Introduction to Deep Agents

Long-running agents and multi-step workflows

Deep AgentsVerify

LangChain Academy

Quickstart: LangSmith Essentials

Tracing, evals, experiments, and agent observability

ObservabilityVerify

Hugging Face

Hugging Face Agents Course

Certificate of Excellence — GAIA multi-tool agent evaluation (~40%)

Certificate of ExcellenceVerify

Great Learning

Prompt Engineering for ChatGPT

Prompt design and LLM prompting

ID: CWUKMBCM

Verified CredentialVerify

Anthropic

Claude Code in Action

Agentic coding workflows with Claude Code

Anthropic VerifiedVerify

Formal Academic Foundation

B.Sc. – Mathematics, Statistics & Computer Science

Krishna Chaitanya Degree College · Andhra Pradesh, India

2018 – 2023

Technical Stack

Technical skills & core disciplines.

Production technologies and agentic frameworks verified across real projects, academic research OS, and live Apple App Store backends.

Core Specialization

Generative & Agentic AI

Architecting multi-agent state machines, cyclical graphs, hybrid retrieval pipelines, and deterministic guardrail gates.

LangGraph
Multi-Agent Orchestration
Hybrid RAG
Tool Use
HITL (Human-in-the-Loop)
Deterministic Guardrails
LangSmith Tracing
Deep Agents

Technologies

LangGraphLangChainChromaDBBM25 / RRFLangSmithOllama

Production Engineering

Backend & Distributed Systems

Building asynchronous REST APIs, authentication layers, stateful context windows, and containerized cloud services.

Asynchronous FastAPI REST API architectures
PostgreSQL and MongoDB data modeling & queries
Stateless JWT authentication & session management
Containerization with Docker & automated PyTest CI
Cloud deployments on GCP & Firebase infrastructure

Technologies

FastAPIPostgreSQLMongoDBFirebaseGoogle Cloud Platform (GCP)Ollama (Local Models)AWS & Azure (Basic)

Engineering Foundations

Languages, Core & Frontend

Deep foundation in Python OOP, mathematical reasoning, type-safe APIs, and responsive React dashboards.

Python (OOP, async/await, Pydantic type safety)
REST API design & OpenAPI specifications
React & Vite for real-time SSE streaming dashboards
Hugging Face Transformers & local model inference
Git version control & structured code review practice

Technologies

PythonREST APIsOOPFastAPITypeScriptReactVite

Approach

How I think.

A disciplined engineering methodology that prizes evidence, observability, and robust failure recovery over superficial demos.

Understand

Define problem & failure constraints

Isolate user requirements, latency tolerances, and data sensitivity before committing to frameworks or model architectures.

Deconstruct core bottlenecks
Specify latency and cost budgets
Identify data boundary risks

Design

Map data flows & state topologies

Draft the explicit state graph, schema definitions, deterministic fallbacks, and observability instrumentation up front.

State machine & transition graph
Schema enforcement (Pydantic / TS)
Degradation & recovery paths

Build

Implement modular, testable units

Code isolated, testable components with defensive error boundaries and transparent telemetry logging.

Decoupled service modules
Deterministic evaluation hooks
Type-safe API contracts

Verify

Stress test & document limitations

Subject the pipeline to adversarial inputs, quantify hallucination rates, and honestly document operational trade-offs.

Adversarial red-teaming tests
Trace span telemetry audit
Explicit limitation disclosures

About Me

The person behind the systems.

AI Engineer specializing in production Generative AI, LangGraph state machines, hybrid retrieval, and Python backend services.

Jeevan Bandanadham

Andhra Pradesh, India

Engineering Background

With a foundational background in Mathematics, Statistics & Computer Science (Krishna Chaitanya Degree College), I approach AI engineering with rigorous quantitative and state-machine thinking rather than superficial wrapper scripting.

My engineering journey spans engineering production agentic backends at Trivexa Technologies, deploying live conversational AI applications on the Apple App Store for WorldTrendz LLC, enterprise GenAI and RAG services at Vaaryan Interfaces, and authoring recognized open-source architectures like Project Anveshan and NyayaPath.

Engineering Principles & Standards

Craft Discipline

Start with the problem, not the framework.

Resist reaching for complex multi-agent graphs when a deterministic heuristic, structured parser, or hybrid search resolves the user's problem with lower latency and zero hallucination risk.

Make agent state transitions observable.

Every node execution, tool call, and state reduction must emit structured trace spans. Full-lifecycle LangSmith instrumentation ensures non-deterministic LLM loops remain debuggable.

Enforce deterministic quality gates.

Generative outputs must pass through strict, zero-LLM sanitizers, statutory allowlists, and factuality assertions before reaching live end users.

Design for failure and regime shifts.

Statutory transitions (like IPC to BNS), API rate limits, and network dropouts will occur. Build date-based temporal routing, fallback triage, and checkpointed state recovery from day one.

Ground claims in empirical test suites.

Measure system performance with automated test suites (like NyayaPath's 140 unit tests) and citation integrity benchmarks, proving reliability over marketing claims.

Field Notes

Technical reflections & observations.

Brief essays and field dispatches documenting non-obvious engineering decisions, failure patterns, and architectural trade-offs.

Observability & Multi-Agent Systems

6 min read

Tracing Agent Workflows: Moving Beyond Naive Demos

Why traditional HTTP request/response logging breaks down in non-deterministic multi-agent graphs, and how trace spans, execution state trees, and token consumption metrics keep production agents debuggable.

Read Note Summary

Information Retrieval & NLP

8 min read

Designing Resilient RAG: Lessons from Hybrid Dense-Sparse Retrieval

Pure vector cosine similarity routinely fails on exact alphanumeric statutory codes, product IDs, and domain acronyms. How combining Reciprocal Rank Fusion (RRF) with cross-encoder rerankers eliminates retrieval blind spots.

Read Note Summary

Local Inference & Edge AI

7 min read

Running Small Language Models on Local Silicon: Practical Constraints

A deep dive into GGUF quantization levels, KV cache memory footprint, and prompt formatting quirks when running Llama 3 / Mistral 8B models locally for private automated workflows.

Read Note Summary

Initiate Contact

Let’s make something useful.

Have a problem worth solving? I’m open to thoughtful conversations about AI applications, backend systems, and high-impact engineering projects.

Direct Correspondence

Jeevanpaul766@gmail.com

Typically responds within 24 hours. Confidential technical discussions welcomed.

Professional Profiles

github.com/Jeevanpaul766 linkedin.com/in/jeevanbandanadham

Send a Direct Message

All messages are routed securely to Jeevan Paul’s inbox.