Bhumika Mahajan

01 — About

A quick intro

AI/ML and software engineering new graduate with hands-on experience building LLM-assisted workflows, recommender-system pipelines, backend services, and scalable data systems. Currently working as an AI/ML Intern at Fox Corporation on Databricks/PySpark-based workflows for personalization, metadata enrichment, embeddings, and content-understanding use cases. Strong foundation in Python, SQL, data structures, algorithms, APIs, machine learning, and production-oriented delivery, with interest in building reliable AI products and scalable software systems that work on real-world data.

5+ML Projects Shipped

10+Skill Domains

FoxCorporation Intern

2026B.E. CSE Grad

02 — Experience

Where I've worked

AI/ML Intern — Fox Corporation

Jan 2, 2026 – Jul 1, 2026

Personalization & Recommendation Team · Databricks Platform

Built and deployed Databricks/PySpark pipelines processing 700K+ media assets (shorts, clips, listings, episodes, movies, specials) for metadata enrichment, semantic retrieval, content discovery, and recommendation workflows.
Cut recompute cost and runtime with a two-tier hash-based CDC layer that reprocessed only changed rows, plus deduplication, caching, and idempotent 2-phase Delta MERGE upserts.
Built LLM-assisted extraction using GPT-5-mini with web search to generate 20+ structured metadata fields (entities, themes, genres, teams/leagues, talent) from ASR transcripts and catalog text.
Generated 3072-dim Gemini embeddings powering similarity, retrieval, and discovery across the catalog, sustaining ~1000 rows/min via a 300-thread concurrent pool and batched API calls.
Hardened reliability with validation checks and batch-level comparisons that caught regressions before production, plus run-log debugging and monitoring on scheduled Databricks workflows.
Collaborated with engineering and product stakeholders to turn ambiguous requirements into reusable, well-documented pipeline components with measurable, reliable outputs.

03 — Projects

Things I've built

DocuMind — RAG Document Q&A

End-to-end Retrieval-Augmented Generation system to chat with your documents. Built the full pipeline from scratch — ingestion, chunking, embeddings, vector search, and grounded answer generation with source citations. Pluggable OpenAI/Gemini/offline backends, FastAPI API + web UI, 16 automated tests, Docker & CI.

PythonFastAPIRAGVector Search

View on GitHub →

ShortLink — URL Shortener + Analytics

Production-style URL shortener (Bit.ly-like) showcasing core system-design concepts — Base62 encoding, LRU caching on the redirect path, token-bucket rate limiting, idempotent shortening, and click analytics. Pluggable in-memory/SQLite storage, FastAPI service, 27 tests across both backends, Docker & CI.

PythonFastAPISystem DesignCaching

View on GitHub →

Emotion-Based Media Recommender

Scalable, component-based responsive UI for four emotion states. Cut redundant code 60%, sped up load times 25%, and raised UX consistency 40% — with integration-ready modules for ML-driven recommendations.

HTMLCSSJavaScriptUI/UX

News Headline Classifier

Fine-tuned a transformer-based BERT model to 92–96% accuracy; optimized preprocessing for 35% faster runs and served a low-latency, REST-ready inference endpoint (<200 ms).

PythonBERTHugging FaceNLP

Image Captioning Model

CNN+LSTM model (BLEU 0.55–0.65) with 30% faster preprocessing; improved fluency via teacher forcing, dropout, and LR scheduling (↓18% perplexity) behind a sub-second pipeline.

PythonTensorFlowCNN+LSTMCV

Multimodal Image, Caption, Story & Audio Generator

Fully automated multimodal pipeline (image → caption → story → audio) at 90% semantic consistency; ↓28% processing time, deployed as a single one-click real-time Gradio app.

BLIPFLAN-T5gTTSMultimodal

Audio CAPTCHA Solver

Hybrid decoding system (MFCC/PCA + multiple models) at 88–94% accuracy with tuned training/inference pipelines, reducing classification error 22% vs baseline.

PythonMFCC/PCASVMConv1D CNN

Want to see more?

Explore my code, notebooks, and experiments on GitHub.

Visit GitHub →

04 — Skills

My toolkit

Programming & CS

PythonSQLC++JavaJavaScript/TSRData Structures & AlgorithmsOOPOperating SystemsDatabasesComplexity AnalysisLinuxGit/GitHub

Backend & Systems

FastAPIREST APIsAPI DesignSystem DesignMicroservicesCaching (LRU)Rate LimitingBase62 EncodingIdempotencyAsync/Parallel ProcessingPython ConcurrencySQLitePytestDockerGitHub Actions

Generative AI, LLMs & NLP

GeminiOpenAI/GPTAzure OpenAIClaudeBERTTransformersHugging FaceFLAN-T5NLP/NLUASR/Transcript ProcessingPrompt EngineeringStructured OutputsTool/Function CallingRAGEmbeddingsLLM Fine-tuningInference OptimizationAgentic AILangChain

Data Platforms & Distributed Systems

Apache SparkPySparkPandas UDFsDatabricksUnity CatalogDelta LakeETL/ELTCDC/Incremental ProcessingDistributed Data ProcessingFeature ExtractionData Wrangling

Databases, Search & Retrieval

MySQLMongoDBVector DatabasesElasticsearchSemantic SearchVector SearchCosine SimilarityTF-IDFRanking & RelevanceLearning-to-Rank

Machine Learning

Supervised LearningUnsupervised LearningXGBoostFeature EngineeringModel EvaluationHyperparameter TuningRecommender SystemsRanking ModelsUser Behavior ModelingCold-start Modeling

Deep Learning, CV & Multimodal

Neural NetworksCNNsRNNs/LSTMsAttentionGenerative AIVision-Language ModelsMultimodal EmbeddingsImage CaptioningOpenCVGrad-CAM/XAI

MLOps, Cloud & DevOps

MLflowUnity CatalogModel ServingMonitoringObservabilityCI/CDDockerKubernetesCloud DeploymentAWSGCP Vertex AIAzure

Frameworks & Tools

PyTorchTensorFlowScikit-learnNumPyPandasMatplotlibKerasGradioFastAPIUvicornStreamlitJupyterGoogle Colab

05 — Education

Academics

B.E. Computer Science

Thapar Institute of Engineering and Technology, Patiala

Expected June 2026 · CGPA 7.7

AI/ML Engineer & Software Engineer