Shubham Bhattacharya — Applied AI Engineer (AI Backend / LLM Applications)

01 / About

Engineering Background

BCA graduate (Dec 2026) at Parul University (CGPA 8.13) building production AI systems with Node.js, TypeScript, PostgreSQL, and pgvector. I focus on RAG architecture, LLM reliability, and AI evaluation — every system I ship comes with measured quality metrics. Based in Vadodara, Gujarat.

Open to Applied AI Engineer Internship roles (AI backend / RAG / LLM systems) where I can ship production code. My engineering philosophy: data integrity, structured logging, and designing for graceful failure.

Zero-Trust Validation

Zod schema validation on every API input and LLM output before database writes.

Multi-Model Resilience

Automatic failover routing across LLM providers for zero downtime during outages.

Observability First

Pino structured logging with Sentry error tracking for end-to-end request tracing.

API-First Design

Typed REST APIs with explicit error handling and clear boundary separation.

02 / Architecture

Backend Architecture Patterns

Patterns I use to build reliable, observable systems.

Model Fallback Routing

Resilient orchestration via Groq SDK. If primary models hit rate limits or fail, automatic failover routes traffic to secondary models to maintain uptime.

Groq SDKLlama 3.3Failover

Zod Schema Validation

Every API request and LLM output is validated against strict Zod schemas. Malformed data is rejected before hitting database logic.

ZodValidationType Safety

Prisma + PostgreSQL

Type-safe queries with Prisma ORM. Relational PostgreSQL schema with migrations, explicit indices, and atomic database transactions.

PrismaPostgreSQLTransactions

Observability Stack

Structured logging with Pino coupled with Sentry exception tracking for end-to-end request tracing and debugging.

SentryPinoLogging

Session Security

Trusted Backend Context isolation pattern preventing prompt injection vulnerability and locking executions into authorized boundaries.

SecurityAuthInjection Shield

CI/CD & Infrastructure

Automated testing pipelines using GitHub Actions, containerized deployments via Docker, and production hosting on Vercel and Render.

GitHub ActionsDockerVercel

03 / Projects

Featured Work

Backend systems I've built, not tutorial follow-alongs.

Backend · AI Infrastructure

RAG Chat — Grounded Document Intelligence

Grounded Document Intelligence — Retrieval‑Augmented Generation over PDF documents with hybrid search, anti‑hallucination grounding, and real‑time streaming. Live Demo: ragchatbot-61jh.onrender.com.

Hybrid SearchCombines vector search (pgvector) with lexical text search (to_tsvector)
Eval Harness14-case golden test set vs live deployment: 100% answer accuracy, 100% grounding, 100% source attribution — including a trap case the model knows but the corpus doesn't (refuses anyway)
Anti-HallucinationDistance gate (MAX_DISTANCE ≤ 0.5): weak evidence never reaches the LLM — refusal over guessing
Token StreamingToken-by-token SSE streaming responses with document attribution

BunFastifyTypeScriptPrismapgvectorGeminiGroqSSE

Source Code Live Demo

Full Stack · AI Auditor

Medo Copilot

AI-powered launch-readiness auditor — built a 3-tier LLM waterfall (Gemini 2.0 → Groq/Llama 4 Scout → OpenRouter/Claude) with automatic provider failover for high availability.

Waterfall Fallback3-tier LLM waterfall with automatic provider failover for high availability
Hybrid ScoringCombines AI analysis with deterministic Lighthouse & static code metrics
Security AuditStatic security analyzer scanning for hardcoded secrets and deployment risks

Next.jsTypeScriptVercel AI SDKClerkGeminiGroq

Source Code Live Demo

Full Stack SaaS · Telemedicine

DocNow

Full-stack telemedicine SaaS platform with symptom triage classifiers, WebRTC consultations, and transaction billing structures.

Symptom TriageSymptom triage classifier routing patients to appropriate specialists
Video CallsBrowser-based WebRTC video consultations via Vonage API
Credit SystemCredit wallet engine with audit trails for purchases, deductions, and refunds

Next.jsTypeScriptClerkPrismaPostgreSQLVonage

Source Code Live Demo

Backend · AI Integration

AI Ops — AI Customer Support Backend

AI customer support backend that enhances support workflows, built with a custom ReAct orchestration engine (no LangChain) for automated order‑tracking and refund draft processing across multi‑step reasoning cycles.

ReAct EngineBuilt a custom ReAct orchestration engine (no LangChain) for automated order-tracking across multi-step reasoning cycles
Tool RegistryDesigned Zod-validated tool registry enforcing schema validation on every LLM output before database writes
Fallback RoutingImplemented Llama 3.3-70B → Llama 3.1-8B fallback routing via Groq SDK for resilience during provider outages

Node.jsTypeScriptExpressGroq SDKPostgreSQLPrismaZod

Source Code Live Demo

04 / Stack

Technical Skills & Technologies

Technologies I use to ship production systems.

Languages

TypeScript JavaScript (Node.js) SQL

Backend & API

Express.js Fastify Bun REST API Design Middleware Architecture SSE / WebSockets Auth (Clerk)

AI & LLM Infrastructure

Custom ReAct Loops Tool Registry Design Multi-Model Routing Prompt Injection Protection tiktoken Cost Tracking

Databases

PostgreSQL Redis Prisma ORM

Observability & Safety

Zod Schema Validation Sentry Error Tracking Pino Structured Logging

CI/CD & Infrastructure

GitHub Actions Docker Compose Vercel Render

05 / Journey

Education & Projects Timeline

BCA — Parul University, Vadodara

2023 — Dec 2026

Bachelor of Computer Applications · CGPA 8.13 · SGPA 8.45 · First Class with Distinction

Relevant Coursework: Soft Computing (AI), Software Engineering, Data Structures & Algorithms, Database Management Systems, Computer Networks, Operating Systems, Object-Oriented Programming

Independent Backend & AI Engineering

2024 — Present

Self-directed system architecture & LLM integration research

Built custom ReAct orchestration engine (no LangChain) for automated order-tracking and refund draft processing
Designed Zod-validated tool registry enforcing schema validation on every LLM output before execution
Implemented multi-model fallback routing (Llama 3.3-70B → Llama 3.1-8B) via Groq SDK for resilience during outages
Developed Trusted Backend Context pattern for prompt injection protection

06 / Contact

Get In Touch

Open to Applied AI Engineer Internship roles. Feel free to reach out through any channel below.

Email shubhambhattacharya107@gmail.com Phone +91-9155252394 GitHub github.com/shubhambhattacharya-dev LinkedIn linkedin.com/in/shubhambhattadev