Open to a 6-month internship (Jan 2027) and SDE / AI-ML roles

I build AI, then prove it works.

I'm Mohammed Aahil Parson, a final-year Computer Science student at PES University. Peer-reviewed anomaly-detection research, a deployed RAG product, and distributed systems built from raw sockets up.

View resume See projects

ais-anomaly-detector — live

From my C3I research: learn what normal vessel tracks look like from unlabelled data, then flag the ones that aren't. Normal trackFlagged

99.2%precision, AIS vessel anomaly detection

94.4%win rate, HMM + DQN Hangman agent (vs 7.18%)

0.955BERTScore F1 keeping a writer's voice

80%+top-3 retrieval hit rate, Contract Risk Analyzer

30+ FPSlive sign-language recognition on webcam

Top 150of ~3,000 teams, Atos Srijan 2026

96thof 1,380 teams, Guidewire DEVTrails 2026

16%acceptance rate at DoSCI-2026, where my paper was published

PyTorchRAG pipelinesFastAPIChromaDBAWS BedrockKafkaSparkDockerReact NativePostgreSQLRedisFAISSscikit-learnC++TypeScriptTCP sockets

Currently building

My final-year capstone, in active development.

In development Team capstone

IntelliHome: house design pipeline

An AI-assisted house-design pipeline exploring conversational requirements, rule-aware planning, floor-plan generation, geometry validation and plan scoring.

My part: engineering, integration, testing and development of the shared system.
Built collaboratively; the canonical repository is owned by a teammate.
Research based on the project is planned; details will be added once formally available.

GitHub (private, access on request)

The repository is private, so implementation details are limited here. Happy to walk through my part in an interview.

Things I've shipped

Each one starts with a real problem and ends with a number.

Contract Risk Analyzer

Live

Legal review is slow because risk has to be found clause by clause. I built and deployed a system that extracts clauses from PDFs, retrieves context from a ChromaDB vector store, scores risk through a multi-provider LLM pipeline and exports a report. Evaluated on 90 benchmark clauses: 80%+ top-3 retrieval, 90%+ contract-type precision, 100% schema validity.

RAG
LLM APIs
ChromaDB
FastAPI
Celery
Redis
Docker

Try it live Code

WriteYou

AI drafts lose the voice of the person signing them. A six-stage retrieve-and-rewrite pipeline over a 44,884-word personal corpus with style embeddings and a 30-signal habit scorer. First author on the paper.

Python
Flask
sentence-transformers

Read paper

Hybrid Hangman AI solver

RL agents guess letters blindly. 24 position-based HMMs feed a Deep Q-Network with a 619-dim state; tested on 2,000 unseen words. Built at an ML course hackathon.

PyTorch
HMM
DQN

Code

Real-time ASL recognition

Fingerspelling read live from a webcam. A custom CNN with a Streamlit app, tuned for an RTX 4060 with mixed precision and Tensor Core-aligned layers. About 85% validation accuracy.

PyTorch
OpenCV
Streamlit
CUDA

Code

Mini HDFS

Files that survive a node failure. I wrote the full Namenode over raw TCP: chunking, placement, replication, heartbeats, reconstruction and hash-verified downloads.

Python
TCP sockets
Multithreading

Code

InsureIt

Parametric travel insurance that pays out without paperwork. I led the Android-first app and the web frontend: policies, claims, payouts and dashboards, with automated UPI payouts.

React Native
Expo
TypeScript
FastAPI

Code

Real-time server monitoring

Catch resource exhaustion as it happens, not in post-mortems. Kafka and Zookeeper route telemetry over ZeroTier to Spark, with alerts for spikes, saturation, possible DDoS and disk thrashing.

Kafka
Spark
Zookeeper

Code

Research

Peer-reviewed work, written up and published.

DoSCI-2026Elsevier SSRN16% acceptance

Hybrid BiGRU autoencoder and One-Class SVM for vessel anomaly detection from AIS data

Co-author. I independently built the pipeline (preprocessing, 10-step trajectory windowing, training and synthetic-anomaly evaluation), reaching 99.20% precision and 84.79% F1.

Read paper Code

IJRPRVol. 7, Issue 6First author

WriteYou: personalised writing-style adaptation engine

A retrieval-and-rewrite pipeline that keeps a person's own voice in AI-drafted text, evaluated at 0.955 mean BERTScore F1.

Read paper

Experience & toolkit

Where I've worked, what I've won, and what I build with.

Experience and wins

AI Research Intern, C3I, PES UniversityJun – Jul 2025. Built the AIS anomaly-detection pipeline end to end; it became a peer-reviewed paper.
Top 150 of ~3,000 teams, Atos Srijan 2026Nimbus, an AI cloud-security copilot.
Copilot backend in ~9 hours, NIMBUS1000AWS Bedrock and FAISS-backed FastAPI backend for a working MVP. Code
96th of 1,380 teams, Guidewire DEVTrails 2026InsureIt, a parametric insurance platform.
B.Tech CSE, PES UniversityExpected June 2027.

Toolkit

Languages

Python
C++
TypeScript
JavaScript
SQL

AI and ML

PyTorch
scikit-learn
CNNs
GRU/LSTM
HMMs
DQN
One-Class SVM
XGBoost

GenAI and retrieval

RAG pipelines
LLM APIs
AWS Bedrock
Embeddings
ChromaDB
FAISS
Eval harnesses

Backend and systems

FastAPI
Flask
PostgreSQL
Redis
Celery
Docker
Kafka
Spark
Linux

Frontend and mobile

React
React Native
Expo
Supabase
Firebase

Hiring an intern or new grad?

Grab the resume that fits the role, or email me at .

SDE resume AI/ML resume