Hardik
Sehgal.
Full-Stack Software Engineer with ~2 years of experience - building business-critical platforms and AI-powered systems.
CRM systems // RAG pipelines // Vector search // LLM integrations
▼ scroll
From concept to production.
Full-stack software engineer with ~2 years of experience, currently an SDE 2 at Omazon Ltd building and maintaining a business-critical CRM - order tracking, returns, warehouse analytics, customer delivery.
Before that: Founding Engineer at GradGuide, building AI-powered platforms and scalable web systems from scratch - system architecture, API design, database schemas, and the product decisions in between.
Core strengths: React.js, Next.js, FastAPI, and modern AI technologies - RAG systems, vector databases, and LLM integrations. Comfortable owning things end-to-end, from concept to production.
- experience
- ~2 years
- currently
- SDE 2 @ Omazon Ltd
- previously
- Founding Engineer @ GradGuide
Case studies, not bullet points.
Each role, told the way it actually happened - the problem, what got built, and what changed.
00
A business-critical CRM driving live operations.
>problem
The business needed a CRM to track and manage order returns, order status, warehouse analytics, and customer delivery - systems with direct, day-to-day business impact.
>built
- ▸Works across the full stack of the CRM - both frontend and backend - for returns processing, order tracking, warehouse analytics dashboards, and delivery status tracking.
>impact
This is the system directly driving live business operations for returns, orders, warehouse, and delivery - not an internal tool or a side project.
01
A student CRM, architected from zero as engineer #1.
>problem
Student data management and counsellor/admin workflows needed a system built from scratch - no existing architecture, schema, or API surface to build on.
>built
- ▸Architected and led full-stack development using FastAPI, Next.js, and TanStack Query.
- ▸Defined the entire system architecture, RESTful API design, and database schema as founding engineer.
- ▸Implemented role-based access control across counsellor and admin roles.
- ▸Designed a media storage pipeline using Google Cloud Storage for documents and Cloudinary for user media.
>impact
Centralized asset management and a working CRM built end-to-end as the first engineer on the product.
02
Semantic discovery over 1600+ German courses.
>problem
Indian students needed a way to discover and compare 1600+ German courses with real relevance - naive keyword search and hard filters alone weren't cutting it.
>built
- ▸Two-stage search architecture combining hard filters (fees, course name, duration) with semantic vector search on Qdrant.
- ▸LLM analysis pipeline for deep course evaluation, powered by Redis RQ async workers processing large scraped university datasets.
- ▸Next.js frontend with TanStack Query for real-time search, comparison, and AI-generated summaries.
>impact
Significantly improved relevance of course recommendations over naive filtering and keyword search.
03
Frontend lead for an AI-powered audit platform.
>problem
An AI-powered auditing platform needed a performant, responsive frontend that could keep pace with the product.
>built
- ▸Led frontend development in React, TypeScript, and Next.js - server-side rendering, dark mode, and mobile-first responsive layouts with Flexbox and Grid.
- ▸Integrated platform APIs using Axios with async/await patterns.
>impact
Improved performance and usability by 25%.
04
Company landing page - auth, documents, CI/CD.
>built
- ▸Built the company landing page backend in Next.js, including user authentication and document submission workflows.
- ▸Set up Docker deployment pipelines and GitHub Actions workflows for automated testing and image builds.
Built after hours.
Academic and personal builds - smaller in scope than the case studies, still designed to work end-to-end.
~/projects/ai-interview-platform
AI Interview Platform
An automated interview system that generates JD-based questions with adaptive follow-ups - voice in, voice out, and a full report at the end.
- ▸Text-to-speech via Sarvam, speech-to-text via Groq
- ▸Anti-cheating: OpenCV face tracking + tab/window-switch detection
- ▸Comprehensive interview summaries - strengths and weaknesses
- ▸Memory graphs with Neo4j + Mem0
Sarvam TTSGroq STTOpenCVNeo4jMem0
~/projects/empower-her
Empower Her
A women's safety app built with Flutter and Firebase - designed around getting help fast, with family in the loop.
- ▸Real-time location sharing
- ▸Parent–child chat interface
- ▸Shake-to-alert emergency trigger
- ▸Integrated emergency services - police and hospitals
FlutterFirebaseBLoCGetX
Learning in public.
Notes I publish while working things out - mostly RAG internals, plus the fundamentals underneath. Every entry opens the full post.
~/writing
$ ls -R ~/writing
rag/
6 postsA series on retrieval-augmented generation and the query-translation patterns that make it work - written in order.
fundamentals/
3 postsThe things underneath the stack - protocols, tooling, and the vocabulary everyone uses before anyone defines it.
$ total 9 · hashnode + medium
$ stack --list languages
PythonJavaScriptTypeScriptJavaDart
$ stack --list frontend
HTML5CSSTailwind CSSReact.jsNext.jsFlutter
$ stack --list backend
Node.jsExpress.jsFastAPIDockerRedis RQLinux
$ stack --list databases
MongoDBQdrantNeo4j
$ stack --list ai-ml
LangChainLangSmithVector SearchRAG SystemsMemory Layer
$ stack --list ci-cd
GitGitHub Actions
B.Tech in Computer Engineering
Bharati Vidyapeeth, College of Engineering, Pune
Apr 2020 - Mar 2024
Query this site.
No LLM behind this prompt - just a static command map, the way a search index would answer. Type 'help' to start.
