AbduRahim Shaikh
Available 16 hrs/week · Nijmegen blazerahim.ar@gmail.com

Builder of things that actually ship

I spent two years building production LLM tools at Vector Consulting Group: auto-documentation, email-to-task routing, semantic search, and document parsing for compliance checks. Now I'm doing an MSc in AI at Radboud, and I'm available 16 hours a week.

See the work Email me
Yanbu → Mumbai → Nijmegen · 10,416 km Scroll to fly the route
Origin
Yanbu, Saudi Arabia24.09°N 38.06°ESchool, 2019–2021
Via
Mumbai, India19.08°N 72.88°EB.Tech, RAIT · Vector Consulting Group
Now
Nijmegen, Netherlands51.81°N 5.84°EMSc AI, Radboud University
Status
Available16 hrs/week alongside the MSc
Yanbu24.09°N 38.06°E · 2019–2021 Grade 12 at Radhwa International School, on Saudi Arabia's Red Sea coast.
Mumbai19.08°N 72.88°E · 3,634 km B.Tech in Computer Engineering (AI) at RAIT, CGPA 8.12. Then two years building production LLM tools at Vector Consulting Group.
6,783 km61.0° of great circle The second leg. A published Springer paper, a second under review, and two years of production work make the trip.
Nijmegen51.81°N 5.84°E · 10,416 km MSc Artificial Intelligence at Radboud University, Machine Learning & Neural Computing. Here since August 2026, available 16 hours a week.

Selected work

Two years of production LLM work, then the research and side projects. Every number here comes from the resume, the paper, or the code.

LLM productivity suite

A modular suite deployed across internal teams at Vector Consulting Group: auto-documentation, email-to-task labelling and speech-to-task conversion. A task-routing module classifies incoming work and assigns it automatically, replacing manual triage. Semantic search runs on Continue.dev, ChromaDB and Azure GPT-4o.
27%lower inference latencyAI Developer · 2025–26

Document intelligence

Qwen2-VL for document parsing and text extraction in internal compliance checks and slide tagging. Quantized LLMs self-hosted on Ollama with ChromaDB vector search. REST APIs in FastAPI, containerised with Docker, deployed through Azure DevOps CI/CD.
35%faster responsesAI Developer Intern · 2024–25

VeriLens

Deepfake video detection. Haar-cascade face crops and VGG16 features read as a sequence by an LRCN, with a calibrated SVM evaluated alongside. Tested on Celeb-DF v2 and DFDC.
0.877AUC · DFDCSpringer MTTA · under review

Hisaab-Book

A multilingual inventory system for small businesses that English-first software leaves out, with voice and image input. Task prediction and usage insights with XGBoost over modular microservices. Python, FastAPI, Supabase, Twilio.
~50%fewer manual entriesBhashini Sprint finalist

DeFogX

Driver assistance that dehazes live dashcam video with an encoder–decoder CNN, then fuses a Dark Channel Prior fog score with temperature, humidity and rain sensors to recommend a safe speed. Built on our own dashcam dataset with synchronised sensor readings across fog, low light, glare and rain.
PyTorch · OpenCVRaspberry Pi

Logistics platform

A Django platform that digitised operations end to end, from gate entry through loading to gate exit, cutting turnaround time and manual effort.
DjangoVector Consulting Group

Published work

I publish the harder stuff as research.

VeriLens: A System for Deepfake Detection using Deep Learning Techniques

Haar cascade face detection, VGG16 spatial features, and a Long-term Recurrent Convolutional Network for temporal sequence classification, with a calibrated SVM evaluated alongside it.

Venue
Multimedia Tools and Applications · Springer
Status
Under review
Data
Celeb-DF v2: 6,229 videos (590 real, 5,639 synthetic); DFDC
Protocol
80/20 video-level split, stratified, seed 42
Test results, per video
Test setVideosSVM AUCLRCN AUC
Celeb-DF v21,2460.8230.790
DFDC6070.8770.739

Both test sets are heavily imbalanced. DFDC carries 14.6 synthetic videos for every real one, so the revision leads with AUC and balanced accuracy rather than raw accuracy, which a classifier can inflate by never predicting the minority class.

VeriFace: Deepfake Detector Using Deep Learning

The first paper, and the groundwork VeriLens builds on. With Rahul Barna, Yashas Kulkarni, Jamal Mydeen and Namita D. Pulgam.

Venue
Communications in Computer and Information Science · Springer
In
Intelligent Computing and Big Data Analytics (ICICBDA 2024), pp. 361–375
Status
Published, 2024
DOI
10.1007/978-3-031-74682-6_24
LLMs & NLP
Azure OpenAI (GPT-4o)OllamaQwen2-VLChromaDBContinue.devSemantic searchText classification
ML & vision
TensorFlowPyTorchscikit-learnXGBoostOpenCV
MLOps & backend
DockerFastAPIDjangoAzure DevOps CI/CDAzure deployment & monitoring
Data
PythonSQLPandasBeautifulSoupPower BIPlotlyA/B testing

How VeriLens
reads a face

Five stages from raw video to a verdict. The points you just flew over are now the network.

  1. Decode15 frames
    Every video is sampled into a 15-frame sequence.
  2. Haar cascade224 × 224 × 3
    Frontal-face detection on each frame, then crop and resize.
  3. VGG16512-d / frame
    ImageNet weights, no classifier head, average-pooled into one vector per face.
  4. Classify15 × 512 → 2
    An LRCN reads the sequence: Dense 256 into LSTM 128. A calibrated RBF SVM is evaluated alongside it, scoring each frame and averaging per video.
  5. VerdictAUC 0.877
    Real or synthetic, per video. The calibrated SVM on DFDC is the strongest result in the revision.

From the training code: SEQUENCE_LENGTH = 15 · VGG16(include_top=False, pooling='avg') · LSTM(128)

Experience,
leg by leg

  1. 2026MSc Artificial Intelligence · Radboud UniversityMachine Learning & Neural Computing, Aug 2026 – 2028. Coursework in Deep Learning, Bayesian Machine Learning and Ethics for AI.
  2. 2026VeriLens under review · Springer MTTAMultimedia Tools and Applications.
  3. 2025AI Developer · Vector Consulting GroupMumbai, Jun 2025 – May 2026. LLM productivity suite, task routing, inference pipelines 27% faster, a Django logistics platform.
  4. 2025B.Tech Computer Engineering (AI Major) · RAITD. Y. Patil Deemed to be University, 2021–2025. CGPA 8.12 / 10.
  5. 2024AI Developer Intern · Vector Consulting GroupMumbai, Jul 2024 – May 2025. Qwen2-VL document parsing, self-hosted LLMs 35% faster, FastAPI and Azure DevOps CI/CD.
  6. 2024VeriFace published · Springer CCISIntelligent Computing and Big Data Analytics, ICICBDA 2024.
  7. 2024Head of Tech Operations · Training & Placement Cell, RAITJun 2024 – Jun 2025. Led the technical teams and modernised infrastructure, reducing downtime.
  8. 2024Bhashini Sprint · finalistThe Government of India's language-AI hackathon, National Language Translation Mission, MeitY.
  9. 2024HackThisFall 4.0 · winnerBest project idea addressing a global problem with creative use of Neurelo. Gandhinagar.
  10. 2023Database Co-Head · Training & Placement Cell, RAITOwned the placement database for 1,000+ students across two academic years.
  11. 2023Solving for India · top 30 of 900 teamsGeeksforGeeks hackathon.
  12. 2022AI Research Intern · AjnaLensMumbai, Aug – Sep 2022. Computer vision and deep learning for task automation in XR.

The short version

I'm an AI developer in Nijmegen, doing an MSc in Artificial Intelligence at Radboud University. Before that I spent two years at Vector Consulting Group in Mumbai building LLM tools that turn unstructured business text into something a team can act on.

I like problems that are real. Not Kaggle real. Not demo real. Production real, where a bug at 2AM matters and someone's business depends on the output.

I went to school in Yanbu, Saudi Arabia, did my engineering at RAIT in Mumbai, and moved to the Netherlands for my master's.

Now
MSc Artificial Intelligence, Radboud University, Nijmegen
Available
16 hours a week alongside the MSc
Before
AI Developer, Vector Consulting Group, Mumbai, 2024–2026
Led
Head of Tech Operations, Training & Placement Cell, RAIT
Degree
B.Tech Computer Engineering (AI), RAIT, CGPA 8.12

Off the clock

Outside of code, I write poetry — the emotionally messy kind. I listen to Joji, Glass Animals, Billie Eilish. I play chess badly and lose gracefully.

And I'm a sucker for anything set in space, Project Hail Mary above all. It is why this site is a planet.

PoetryChessIndie sci-fi Atmospheric musicGaming

Tell me what
you are building

I'm in Nijmegen and available 16 hours a week alongside my MSc. If it involves models that have to work outside a notebook, I want to hear about it.

blazerahim.ar@gmail.com LinkedIn GitHub VeriFace on Springer