Backend Engineer& AI Engineer

2+ years architecting scalable cloud-native B2B SaaS platforms.

Specialized in distributed backend systems and enterprise GenAI architectures.

VIEW SELECTED WORK→

RESUME (PDF) →

01 / EXPERIENCE

Full-Time Engineering

PROFESSIONAL TIMELINE

Software Engineer

Full-time

VakilDesk Technologies Pvt. Ltd. · Remote

OCT 2024 — OCT 2026

Spearheaded core backend architecture, AWS cloud infrastructure, and enterprise GenAI systems for mission-critical B2B legal tech SaaS platforms.

Spearheaded end-to-end architecture and technical requirements for 4 enterprise B2B SaaS platforms using React and Django, delivering 100% on-time across 10+ corporate clients.

Engineered high-availability cloud infrastructure on AWS (EC2, S3, RDS) with Docker and Jenkins CI/CD—cutting deployment cycles by 35% with 99.9% SLA uptime, and reducing hosting costs by 50% via memory leak fixes and EC2 rightsizing (32 GiB to 8 GiB RAM).

Pioneered production GenAI RAG pipelines leveraging Gemini API, LangChain, and vector embeddings for semantic retrieval, reducing automated legal drafting turnaround time by 60%.

Orchestrated fault-tolerant asynchronous microservice workflows across DocuSign, eMudhra, GST, and WhiteBooks APIs via Celery, processing 10,000+ monthly transactions with zero data loss.

Optimized high-throughput PostgreSQL databases by refactoring complex Django ORM queries and implementing Redis caching layers, eliminating N+1 bottlenecks and slashing latency by 40%.

Developed WebDAV real-time document collaboration workflows for live MS Word editing and PDF rendering across 5,000+ legal documents, backed by automated Pytest suites (85%+ coverage) and multi-tenant security (JWT, OAuth2, RBAC).

CORE TECHNOLOGIES

PythonDjangoFastAPIPostgreSQLAWS (EC2, S3, RDS)DockerCeleryRedisGemini APILangChainJenkins CI/CDReact

02 / SELECTED WORK

Production Software Products

ARCHITECTURES & SYSTEMS

PROJECT 01 — DOCUMENT INTELLIGENCE PLATFORM

Docvyn

GITHUB REPO

Enterprise multi-tenant document intelligence and Retrieval-Augmented Generation (RAG) platform. Enables non-blocking asynchronous processing of complex business documents with verifiable page-level source citations.

pgvector & Dense Search

Indexes 384-dimensional FastEmbed vectors in Supabase PostgreSQL with strict SQL-level tenant isolation.

Sub-Second Inference & Fallback

High-throughput generation powered by Groq LPUs (Llama 3.3 70B) with automatic Gemini API fallback and zero embedding cost.

Prompt Injection Defense

Rigid <DOCUMENT_CONTEXT> boundary defense mitigating prompt exploits from unvetted enterprise documents.

Automated Verification

Covered by 48 automated unit and integration tests across Pytest, HTTPX, and Vitest.

ENGINEERING STACK

FastAPIReact 18PostgreSQL (pgvector)SupabaseCeleryRedisGroq LPUGemini APIFastEmbedSQLAlchemy 2.0Pytest

docvyn.internal — RAG Citation Workspace

USER PROMPT / DOCUMENT QUERY

"What are the indemnification obligations and liability caps under Section 8.2?"

VECTOR MATCH: pgvector (cosine: 0.182)Page 14 · Chunk 42

"...Each party shall defend, indemnify, and hold harmless the other from liabilities, damages, and costs arising out of intellectual property claims, capped at 2x annual contract value..."

AI SYNTHESIS — GROQ (LLAMA 3.3 70B)380ms · Grounded

Under Section 8.2, indemnification is mutual for third-party intellectual property infringement claims. Aggregate liability is explicitly capped at twice the total fees paid within the preceding 12-month period.

[Source 1: master_services_agreement.pdf · Page 14]

PostgreSQL 16 + pgvector

Docvyn Engine v1.2

PROJECT 02 — WORKFLOW ORCHESTRATION & AGENTIC SYSTEMS

FlowForge AI

GITHUB REPO

Enterprise visual DAG workflow orchestration platform that designs, executes, and benchmarks AI workflows with durable state persistence, distributed worker concurrency, and human-in-the-loop governance.

Topological DAG Scheduler

Custom Kahn's algorithm scheduler with parallel step execution, cycle detection, and SHA-256 idempotency guaranteeing 100% durable state execution.

Sub-40ms Asynchronous Dispatch

High-throughput task queue orchestrated via Upstash Redis and distributed Celery workers with zombie lease recovery.

Deterministic LLM Guardrails

Pydantic schema validation, role-gated human review approval queues, and automated hallucination benchmarks.

Exhaustive Verification

Validated by 62 unit and integration tests across distributed workers and API routes at a 100% pass rate.

ENGINEERING STACK

FastAPIReact FlowPostgreSQL 17CeleryUpstash RedisSupabaseGroq LPUsOpenAIPydantic v2DockerPytest

flowforge.internal — DAG Execution Canvas

STEP 01 · DONE

Document Ingestion

PDF payload parsing & chunk validation

STEP 02 · ACTIVE

Semantic Classifier

Pydantic schema gate via Groq LPU

STEP 03 · PENDING

Human Review Gate

Role-gated review > $50 threshold

KAHN'S ALGORITHM TOPOLOGICAL STATEDispatch: 28ms · In-Flight: 1

> in_degree: [Node_A: 0, Node_B: 1, Node_C: 2]

> idempotency_key: sha256("wf_7721")

> distributed_lease: worker-03 (healthy)

Topological Engine Active

FlowForge Core v2.4

03 / TECHNICAL STACK

Core Disciplines & Tooling

SYSTEM ARCHITECTURE

01 — LANGUAGES & RUNTIMES

Core languages utilized for backend services, scripting, data manipulation, and web clients.

PRIMARY FOCUS

PythonSQL (PostgreSQL, MySQL)JavaScriptTypeScript

EXTENSIONS & PRACTICES

HTML5CSS3Bash / Shell

02 — BACKEND ENGINEERING & ARCHITECTURE

Frameworks and distributed patterns for high-throughput, low-latency microservices.

PRIMARY FOCUS

FastAPIDjangoDjango REST FrameworkCeleryRESTful MicroservicesSystem DesignRedis Caching

EXTENSIONS & PRACTICES

Asynchronous ArchitectureEvent-Driven WorkflowsORM Query TuningWebSocketsWebDAV

03 — AI & GENERATIVE AI ARCHITECTURES

Enterprise-grade RAG pipelines, dense embeddings, vector retrieval, and deterministic LLM guardrails.

PRIMARY FOCUS

Gemini APILangChainLangGraphRAG PipelinespgvectorFastEmbedGroq LPUs (Llama 3.3 70B)Prompt Orchestration

EXTENSIONS & PRACTICES

Vector EmbeddingsPydantic Schema ValidationEvaluation SuitesMulti-Tenant Vector Isolation

04 — DATABASES & CACHING

Relational data modeling, vector extensions, indexing, and distributed cache layers.

PRIMARY FOCUS

PostgreSQLpgvectorSupabaseRedisSQLAlchemy 2.0

EXTENSIONS & PRACTICES

MySQLUpstash RedisDatabase IndexingTransaction IsolationAlembic Migrations

05 — CLOUD & DEVOPS INFRASTRUCTURE

Containerized deployment pipelines, cloud provisioning, and resilient server infrastructure.

PRIMARY FOCUS

AWS (EC2, S3, RDS, IAM)DockerLinux / UnixJenkins CI/CDNginx

EXTENSIONS & PRACTICES

Docker ComposeServerless DBsCost & RAM RightsizingUptime & SLA Monitoring

06 — FRONTEND ENGINEERING

Component architecture, interactive flow graphs, and modern web applications.

PRIMARY FOCUS

React 18React FlowNext.jsTailwind CSS

EXTENSIONS & PRACTICES

Responsive DesignState ManagementViteClient Portal Architecture

07 — TESTING, SECURITY & INTEGRATIONS

Automated test suites, authentication standards, and third-party enterprise service integrations.

PRIMARY FOCUS

PytestJWT & OAuth2RBAC SecurityThird-Party APIs (DocuSign, eMudhra)Git & GitHubPostman

EXTENSIONS & PRACTICES

VitestHTTPXCORS / Rate LimitingeSignature WorkflowsGST / Tax APIs

04 / ABOUT

Background & Philosophy

ENGINEERING PRINCIPLES

I am a software engineer focused on building robust backend systems, resilient distributed data pipelines, and production-grade Generative AI applications.

Over the past 2+ years, I have architected and deployed multi-tenant B2B SaaS platforms serving enterprise clients. My core work centers on solving concrete engineering challenges: eliminating database bottlenecks, designing deterministic workflow engines, cutting infrastructure overhead, and orchestrating fault-tolerant asynchronous microservices.

In Generative AI, my emphasis is on determinism and reliability. Moving beyond exploratory chatbot scripts, I design and build production-grade architectures—such as grounded RAG pipelines with verifiable page citations (Docvyn) and topological DAG workflow orchestrators with automated schema repair and human approval gates (FlowForge AI).

I believe that good software engineering is defined by clear architectural boundaries, strict data isolation, comprehensive automated testing, and disciplined system design.

EDUCATION

Thadomal Shahani Engineering College

B.E. in Electronics & Telecommunications

CGPA: 8.82 / 10.00·Class of 2024

Coursework: Data Structures, Analysis of Algorithms, Object-Oriented Programming, Database Management Systems.

ENGINEERING PILLARS

Deterministic AI & Strict Guardrails
Zero-Data-Loss Asynchronous Queues
Multi-Tenant Data Isolation in SQL
High Test Coverage (Pytest & Vitest)

05 / CONTACT

Initiate Conversation

DIRECT REACH

HAVE A PROJECT IN MIND?LET'S TALK.

Whether you are looking to architect a scalable backend platform, build production-grade GenAI pipelines, or discuss engineering roles—reach out directly.

roshany.ir.ry@gmail.com +91 9834905805

LINKEDIN GITHUB DOWNLOAD RESUME (PDF)