Available for new projects

I build AI tools you can actually click.

Phone agents that book rooms, search that finds the clause in a 200-page PDF, and inboxes that sort themselves. Five products. Four of them are live right now.

Currently taking on 1–2 select client builds — let's talk scope.

Karan Chandel
Karan Chandel
AI engineer · ships full products
5
Products shipped
4
Live right now
1 day
Typical reply
About

Who you’re hiring

The products below are mine. Here’s the person behind them.

I’m Karan Chandel, an AI engineer who ships full products — not prototypes.

I turn manual workflows — case file searches, invoice entry, missed calls, email triage — into software people actually use. UI, API, deployment, all done by me. That’s why two of the five products here have a Live link you can open today.

Python, Django/FastAPI, LangChain/LangGraph, Twilio, Pinecone — whatever the product needs. 5 products shipped solo, 2.5+ years building production AI systems. No logo wall yet, just proof it works.

Karan Chandel
RoleAI engineer
Experience2.5+ years, Python / production AI
I buildVoice agents, RAG, agentic workflows
StackDjango, FastAPI, LangChain
Selected work

Five products. Two live.

Each one is a real UI, a real API, and workers doing the heavy lifting. Open a live demo where one is up, or watch a short walkthrough.

Lexora legal AI case management dashboard showing case list and chat panel
Legal Tech

Lexora — Legal AI Case Management

A workspace for law firms to manage cases, find answers in their own documents, and keep hearings in one place.

Result Cuts case-file review from hours of re-reading to a cited answer in seconds.

Problems it solves

  • A lawyer needs a clause from a 200-page PDF and cannot remember which file it was in
  • Associates spend hours re-reading the same case files the night before a hearing
  • Generic AI answers cannot be used with a client because there is no source to point to
  • Hearing dates live in notebooks or WhatsApp chats and get missed
  • Cases, documents, and notes are scattered across folders, email, and spreadsheets
  • A new associate joining a matter has no fast way to ask the file “what happened so far?”

Technical highlights

  • PDF upload pipeline: Django saves the file, Celery chunks and embeds, Pinecone stores vectors by case
  • Case-aware RAG chat on FastAPI + LangGraph + Groq, with answers tied back to source passages
  • Docker Compose stack deployed on AWS EC2
Django FastAPI Pinecone Groq

In development testing: 40+ uploaded case PDFs across 3 practice areas — response time under 4s per query. Built and deployed solo — from schema design to AWS deployment — over several weeks.

Deskline homepage — hospitality voice operations landing page
Hospitality

Deskline — AI Voice Agents for Hospitality

AI answers the hotel or restaurant phone, books a room or a table, and logs the call into an ops console.

Result Takes the missed front-desk call and writes the booking — with transcripts, tools, and analytics.

Problems it solves

  • Front desk misses calls when the lobby is busy
  • Guests hang up instead of waiting on hold to book a room or a table
  • After-hours bookings go to voicemail and never get followed up
  • Staff retype the same reservation details into the system after every call
  • There is no record of what the caller asked or whether a booking was made
  • Hotels need a demo-ready console clients can trust on sight

Technical highlights

  • Django + DRF + JWT ops console with Retell voice agents and tool webhooks
  • Room and table booking tools, call history, analytics, and browser test calls
  • Docker Compose stack deployed on AWS EC2
Django Retell Groq

Walkthrough of the ops console, booking tools, and call transcripts.

SupportAI admin dashboard showing classified support tickets and agent queue
Customer Support

SupportAI — Customer Support Automation

Inbound emails become classified, prioritized, and assigned tickets — with a separate queue for admins and agents.

Result Stops a shared inbox sitting unread for hours — mail arrives already classified, prioritized, and assigned.

Problems it solves

  • Support emails sit unread for hours because nobody triaged the shared inbox
  • An angry VIP looks the same as a password-reset in a pile of unsorted mail
  • Agents cherry-pick easy tickets and SLA on the hard ones slips
  • Managers cannot see volume, sentiment, or who owns what without a spreadsheet
  • The same “how do I reset…” reply gets typed out by hand every day
  • Weekend mail piles up until Monday with no routing and no owner

Technical highlights

  • IMAP poll of a shared support inbox creates or matches a customer and opens a ticket
  • Groq classifies category, priority, and sentiment, with a rule-based fallback
  • Admins see the full workspace; agents only see tickets in My Queue
Django IMAP Groq
Autonomous AI Strategy Engine report with seven parallel agent outputs
Multi-agent AI

Autonomous AI Strategy Engine

Give it a business question. Seven agents research in parallel and hand back one strategy report.

Result Turns weeks of tab-and-doc research into one strategy report from 7 agents in a single run.

Problems it solves

  • A founder spends weeks researching a market move and still has no single report to share
  • Competitive notes live in five browser tabs and a messy Google Doc
  • Strategy decks skip risk because nobody had time to look it up
  • The team debates expansion with opinions instead of sources
  • Uploaded research PDFs never get used when the actual plan is written
  • Waiting on a consultant for a first-pass strategy that should take one afternoon

Technical highlights

  • 7 parallel agents covering market, competitors, risk, monetization, and roadmap
  • Hybrid RAG over your documents with Cohere + Pinecone
  • Live progress streamed to the UI via SSE
LangChain Groq Pinecone

In development testing: 7 agents in one run — report streamed via SSE.

Fieldline document intelligence dashboard with PDF upload and extracted invoice fields
Document AI

Fieldline — Document Intelligence

Upload PDFs, extract the fields, check them on the page, then export JSON or CSV — without retyping invoices.

Result Cuts invoice entry from retyping every field to extract, check on the page, export.

Problems it solves

  • Someone retypes every invoice into a spreadsheet by hand
  • A wrong amount gets copied and nobody notices until the client calls
  • Different document layouts break a one-size-fits-all template
  • Ops cannot tell which files failed versus which still need a human look
  • Extracted data has no way back to the PDF page it came from
  • Shipping JSON to another system means another round of copy-paste

Technical highlights

  • Celery renders each page with PyMuPDF and captures word boxes
  • Groq returns structured JSON, then values are matched onto the page
  • Human review UI with a live page image and highlighted matches
PyMuPDF Celery Groq

In development testing: sample invoices, receipts, and freight bills — extract, review on the page, then export.

And more

Got a messy workflow?

Tell me what your team does by hand. I’ll tell you what’s realistic to automate — and what isn’t.

The pattern

Different problem. Same kind of build.

Messy input in. Something a human can trust out. That’s the job — whether it’s a PDF, a phone call, or an inbox.

What they said
What it became
"We miss calls when the front desk is busy."
→ A voice agent that answers, books, and logs the call Deskline
"Every market decision takes weeks of research."
→ Seven agents research in parallel and write one report Strategy Engine
"We can't find anything in our case files."
→ Search the files and get answers with sources Lexora
"Someone retypes every invoice by hand."
→ Extract the fields, check them on the page, export Fieldline
"Support emails sit unread for hours."
→ Email becomes a classified, assigned ticket SupportAI
"Ours is ______________."
→ Tell me. I’ll say what’s realistic →
How I can help

Usually it starts as one of these

If your problem doesn’t fit a box, that’s normal. Most real ones don’t.

Chat with your documents

Ask a question. Get an answer with a source you can open.

Pull data out of PDFs

Invoices, forms, packets — extracted, checked on the page, exported.

Answer the phone or the inbox

Bookings on a call, or tickets from email, without a person on every one.

Working on something that doesn't fit neatly into these? Good — most real problems don't. Tell me about it.

Ready to talk? This is the fastest way to reach me.

Got a workflow worth automating?

Send a few lines about what your team does by hand.

I reply within a day.

Prefer to skip the form? Email · WhatsApp

Need something to forward internally? Download a one-page PDF.

I reply within a day.

Prefer email or WhatsApp instead? Email · WhatsApp