OPEN_TO_WORK = True

Anish
Kumar_

ML Engineer in Training / BTech AIML Student

python3 — anish_kumar.py

View Projects → Get in Touch

About

Model Overview

Experience

3+ yrs

Projects

Frameworks

Inference

80ms

recruiter_summary.json

✓ Computer Vision

✓ NLP Systems

✓ RAG Pipelines

✓ FastAPI Deployment

✓ ML APIs

✓ PyTorch + TensorFlow

I'm a BTech AIML student focused on building real-world machine learning systems with emphasis on deployment, inference optimization, and production-style workflows.

My work spans computer vision, NLP, RAG systems, and predictive modeling. I enjoy turning ML concepts into practical applications using modern tooling across training, deployment, APIs, and frontend integration.

Currently exploring: LLM serving, vector databases, CUDA optimization, and scalable inference systems.

TRAINING_CURVE — anish_growth.log

train_loss

val_acc

Skills

Proficiency Radar

Activity

GitHub Stats

…

Repositories

…

Total Stars

…

Followers

Projects

Model Registry

// hover cards to flip and see full details →

computer_vision

real_time_object_detection.py

YOLO-based detection pipeline with live video. Custom-trained on domain dataset, edge-optimized.

mAP

94.2%

FPS

Dataset

12K

real_time_object_detection.py

YOLOv8 fine-tuned on a custom dataset, served via FastAPI at 45FPS on edge hardware.

camera → preprocess → YOLOv8 → FastAPI → frontend

PyTorchYOLOv8OpenCVFastAPIDocker

nlp

sentiment_classifier_bert.py

Fine-tuned BERT for multi-class sentiment. REST API handling high-throughput production traffic.

Accuracy

91%

RPS

500+

Classes

sentiment_classifier_bert.py

Fine-tuned BERT on 80K reviews, deployed with FastAPI + Docker handling 500+ requests/sec.

raw text → tokenizer → BERT → softmax → REST API

TransformersBERTFastAPIDocker

ml_classic

churn_prediction_pipeline.py

End-to-end churn prediction with feature engineering and stakeholder dashboard.

Precision

87%

Recall

83%

Rows

500K

churn_prediction_pipeline.py

End-to-end pipeline with SMOTE balancing and a Streamlit dashboard — reduced churn by 18%.

raw data → feature eng → SMOTE → XGBoost → dashboard

XGBoostScikit-learnStreamlitPandas

computer_vision

face_recognition_attendance.py

Automated attendance via FaceNet embeddings with live camera for 200+ users.

Accuracy

98.4%

Latency

80ms

Users

200+

face_recognition_attendance.py

FaceNet + cosine similarity for live attendance — 98.4% accuracy serving 200+ users at 80ms.

camera → FaceNet → embeddings → cosine sim → SQLite log

TensorFlowFaceNetOpenCVSQLite

nlp

document_qa_rag.py

RAG pipeline for multi-document Q&A using transformer embeddings + FAISS vector search.

BLEU

0.81

Docs

1K+

Latency

1.2s

document_qa_rag.py

PDF ingestion into FAISS vector store, retrieved context fed to LLM for grounded answers. BLEU 0.81.

PDF → chunker → embeddings → FAISS → LLM → answer

LangChainFAISSHuggingFacePython

ml_classic

fraud_detection_xgb.py

Imbalanced classification for credit card fraud. SMOTE + ensemble on 2M records.

Recall

99.2%

0.94

Records

fraud_detection_xgb.py

SMOTE + cost-sensitive XGBoost on 2M records with MLflow tracking — 99.2% minority class recall.

transactions → SMOTE → XGBoost → MLflow → scoring API

XGBoostSMOTEMLflowScikit-learn

Experience

Training History

2023 — present

BTech — AI & Machine Learning

[ Lovely Professional University ]

Core coursework: Deep Learning, Computer Vision, NLP, Data Structures, Statistics. Building and deploying ML systems as major projects alongside academic work.

2023 — 2026

Independent ML Projects

Self-Directed

Built 6 end-to-end ML projects spanning CV, NLP, and predictive modeling. Focused on deployment-quality code, inference optimization, and production-style architecture.

Education

Checkpoints

B.Tech — AI & Machine Learning

[ Lovely Professional University ]

Expected: 2027

Certifications

deeplearning.ai / Coursera / Google

Ongoing

Systems

ML Systems I Can Build

Inference APIs

FastAPI inference pipelines with model serving and REST endpoints.

RAG Pipelines

Vector search, embeddings, chunking, retrieval, and grounded generation.

Computer Vision Systems

Detection, recognition, tracking, preprocessing, and live inference.

ML Dashboards

Interactive dashboards using Streamlit and deployment-ready APIs.

Now

currently_learning/

→ CUDA optimization

→ RAG evaluation pipelines

→ LLM inference serving

→ Vector databases

→ Efficient transformer deployment

v2.1 → Added deployment-focused project architecture

v2.2 → Improved recruiter mode + reduced motion

v2.3 → Added ML systems engineering section

model.fit(next_role)

Open to full-time ML Engineer positions. Responds within 24hrs.

↓ Download Resume Email GitHub LinkedIn

AnishKumar_