01
Websites & internal tools
Business websites, landing pages, dashboards and full-stack apps that are responsive, easy to navigate, and connected to the APIs or data they need.
Full-stack builds
Websites · Applied AI · Business automation
I help small businesses and lean teams launch useful websites, internal tools, AI assistants and workflow automations — especially when repetitive admin is slowing them down or a manual process needs a better system. I can take a project from a clear first scope through the interface, backend, integrations, testing and deployment.
Services
Start with the business problem, not the technology. Tell me what you want to launch or what is taking too much time today. I’ll help define a practical first version, explain the trade-offs, and build only what the project actually needs.
Business websites, landing pages, dashboards and full-stack apps that are responsive, easy to navigate, and connected to the APIs or data they need.
AI document search and assistants that answer from your source material, show citations, and make uncertainty clear instead of pretending every answer is correct.
Connect forms, lead capture, email, spreadsheets and AI steps to reduce repetitive admin and make follow-up easier to track.
Experience
RLHF data collection and model evaluation. Evaluated and ranked 50+ model responses a day and identified failure modes that fed into production model updates.
Built automated testing pipelines in Python and PowerShell.
Remote. Ran unit tests and contributed to code reviews.
Yashoda Technical Campus, Satara (affiliated to DBATU). Final-year project: a MERN-based Smart Municipal Management System, built and deployed independently.
Selected work
These examples show how I approach product problems: what the user needs, how the workflow works, and which parts I built. Some are independent demos or concepts, not paid client case studies; I label them honestly so you can judge the work itself.
Upload PDFs, docs or notes. Groundwork searches by keyword and by meaning, re-ranks the results, then writes an answer that cites where it came from. When free hosting kept wiping the search index, I moved storage to Qdrant Cloud so it stays online.
Ragas evaluation run — 1.00 faithfulness · 0.71 answer relevancy on the evaluated sample.
Instead of one oversized model response, the review runs as coordinated steps: triage, review, and escalation of risky changes (like auth or payments) to a human. It comes with a React dashboard, sign-in and live monitoring.
Every claim in the draft is checked against your source notes. If the faithfulness score falls under 0.85, the report is blocked instead of being sent on. Open source, with an MCP connection to SuperDocs.
Guests browse retreats, send inquiries and chat with an AI assistant. Every inquiry lands in a lead-management view, so the owner can follow up and nothing gets missed.
The workflow takes a new lead, drafts a follow-up with an AI step, sends it through Gmail, and logs the activity in Google Sheets. It demonstrates how n8n, Groq, Gmail, and Sheets can work together; real deployments should be tested against the business's consent and follow-up rules.
An early product concept for roofing contractors: bring missed calls and website enquiries into one workflow, qualify the request, offer an inspection booking, and keep the lead ready for follow-up. This is a concept demonstration, not a live customer deployment or a claim of measured business results.
I contributed to a preschool website built by a small team under a freelancer, and owned four front-end features: a branded navigation bar with the school logo; a programs page where choosing a program opens the admission form with that program already selected, so parents have less to fill in; an animated home-page headline that rotates key messages; and an image carousel. Other team members built the rest of the site.
Have a website, AI workflow or internal tool in mind? Book a short call and we can identify a practical first step for your project.
Book a project call ↗From first call to final ship
Same view, different question at every stage. Scroll to move from what the business needs to what actually gets shipped — discovery, design, engineering, polish, launch.
Understand the business, the audience, the problem and what the finished experience must change.
Questions before we start
Python is my strongest language. For AI work I use FastAPI, LangGraph, vector search (Qdrant) and evaluation tooling (Ragas, Langfuse). On the product side I use React, TypeScript and Next.js.
Yes. Groundwork, AI CodeReview Pro and NotesToReport were each built solo, from the backend and AI layer to the React interface and deployment.
I worked on LLM evaluation and RLHF data: ranking and reviewing model responses and documenting failure modes. My own projects carry that into practice. Groundwork reports Ragas scores, and NotesToReport blocks a report when its faithfulness score is below 0.85.
Websites and internal tools, AI assistants and document search, and business workflow automations. I’m especially useful when a project needs both the user-facing interface and the backend or integrations behind it.
Yes. I can work directly with a founder or alongside an existing team. We can start with a clearly scoped first version, then decide what is worth building next.
Yes. I can work inside an existing repository, design system, API surface, or delivery process and contribute without forcing the team into a completely new stack.
Yes. I’m remote-first, based in Satara, India, and open to Pune. I’ve worked remotely before and I’m used to owning a task from start to finish.
A few lines are enough: what your business does, what you want to build or improve, what happens today, and your ideal timeline. If you know your budget range, include it; if not, we can first discuss the scope.
Contact
Book a short call to explain your business, what you want to improve, and what happens today. If you know your timeline and budget, bring those details too — no polished brief is needed.