Jamil Khan — Agentic Developer

01 / Projects

Selected work

Open source · Legal data · v1.14

JITS Legal

Indian court judgments, turned into clean, searchable data. JITS reads each judgment and pulls out the court, date, cited cases and the laws it invokes, including the 2024 switch from IPC/CrPC to BNS/BNSS.

There's no AI guessing involved. Every field comes from a written rule, so running it twice gives exactly the same result.

DOI10.57967/hf/9847

14,678Judgments processed
1,500+Hugging Face downloads
18.97MSimilarity edges
0Duplicate IDs

Python
NLP
Citation graphs

Code ↗ Dataset ↗

Research · Particle physics

Hidden Vertex

The Large Hadron Collider throws away almost all of its collision data, keeping only events that match what physicists expect to see. Anything truly new could be lost.

Hidden Vertex trains a graph neural network on ordinary collisions only, so anything it can't explain stands out. Part 1 showed that looking at particles one by one isn't enough to spot new physics, which sets up Part 2.

1.1MCollision events
429Model parameters
Part 1Complete

PyTorch Geometric
Graph networks
Anomaly detection

Code ↗ Dataset ↗

Winner · TechXpression 2025

Sentinel

Real-time fraud detection for UPI and bank payments. Two fast rule checks catch the obvious attacks first; a neural network only looks at what gets past them. That keeps a rare but genuine purchase from being flagged as fraud.

Every payment, three checks

Too fast? blocks money mules in under 1 ms
Too much? blocks split-up transfers
Too unusual? the neural network decides
Approve or verify

TensorFlow
VAE
Real-time

Code ↗

Winner · Apify Partner Program, Hack This Fall 2025

Playlistify AI

Describe a mood in English, Hindi, Tamil or Telugu, and get a Spotify playlist that fits. It's packaged as a tool other AI agents can call directly.

From a sentence to a playlist

Your words in four languages
Read the mood 50+ emotional concepts
Rank tracks via the Spotify API
Playlist

NLP
MCP
Docker

Code ↗ Try it ↗

Along the way, three hackathon wins. Mostly curiosity that happened to go well.

02 / Capabilities

What I work with, and where to see it

SkillWhere I've used itProof
Rule-based NLP Extracting citations and laws from 14,678 judgments JITS
Graph neural networks Collisions as graphs, anomaly detection without labels Hidden Vertex
Deep learning A variational autoencoder for rare-event detection Sentinel
Agents & MCP A tool other AI agents can call Playlistify
Open data A public dataset with a citable DOI JITS

03 / About

Hi, I'm Jamil.

I'm a Computer Engineering student who likes building things that hold up under scrutiny.

Most of my work circles one question: when a system hands you an answer, how do you know it's right? Producing answers has become cheap. Checking them, and knowing what to throw away, is where the real thinking happens. That's what draws me to agent systems: they give software room for a second opinion.

Away from the keyboard, I play chess, read fiction, and watch other people play horror games. Actually playing them is a dream I've been forced to postpone until retirement.

StudyingB.E. Computer Engineering
FocusAgent systems and evaluation
AwardHack This Fall 2025, Apify winner
AwardTechXpression 2025, winner
ElsewhereGitHub · Hugging Face · Journal

04 / Writing

I also write to think.

Devlith is my journal: what I'm building, what I'm learning, and what I'm thinking along the way.

Read the journal →

Latest entries

All entries →

05 / Contact

Open to agentic AI internships.

Email is fastest. I reply to anything specific.

khanjamilahmed202@gmail.com LinkedIn ↗ GitHub ↗ Résumé ↗