Maanik Garg — AI/ML Engineer | Frankfurt, Germany

About Me

Making AI work
where it matters most

AI/ML Engineer based in Frankfurt am Main, with a Master's in Data Science and 4+ years building machine learning solutions across production environments. I've spent my career at the intersection of ML and product engineering, building systems that don't just demo well but hold up in production under real traffic.

At WAMOCON, I lead the AI engineering effort: designing intelligent assistants that resolve support tickets without human intervention, building agentic workflows that replaced entire manual processes, and shipping full-stack platforms used daily by trainers and trainees across Germany. I care about clean architecture, fast iteration, and results you can measure in saved hours and reduced costs.

Years in Production

70%

Faster Workflows

90%

Tickets Resolved by AI

Output per Sprint

neural network
data pipeline
inference

Career Path

Where I've delivered impact

AI Engineer

Aug 2024 — Present

WAMOCON GmbH, Frankfurt, Germany

Built an autonomous marketing system using LangGraph and Gemini 2.5 Pro that generates weekly campaign calendars, daily topics, and posting schedules automatically. Cut the team's strategy planning time from days to hours with sub-200ms response times through Redis semantic caching.
Designed a dual-pipeline content workflow that separates generation from quality review, using Gemini 2.5 Pro with Vertex AI Guardrails to catch hallucinations, and Jira + Google Sheets integration to track state. Tripled the marketing team's content output while keeping a human-in-the-loop gate for 100% educational accuracy.
Led the architecture of the FIAE Learning Platform (Next.js + Supabase + Drizzle ORM), deployed via Docker on Strato. A multi-role system with RBAC serving trainers, trainees, and management across three vocational programs (FIAE, KfB, KI Manager), used daily across Germany.
Created "HAI," the platform's AI assistant, with a three-tier query architecture: Page Indexing for exact PDF retrieval, PGVector with fine-tuned embeddings and RAG for semantic queries, and SQL for deterministic UI operations. Dynamic routing sends each query to the right retrieval strategy. Result: 90%+ of routine support tickets handled automatically, taking real pressure off trainers.
Set up the full MLOps lifecycle including drift detection and automated testing, so models stay reliable without constant babysitting.

Software & Data Associate

Dec 2020 — Feb 2024

Cognizant, Pune, India

Built ML models (Random Forest, XGBoost) for semiconductor wafer fault detection, handling severe class imbalance with SMOTE and advanced sampling. Reached 87% accuracy validated through k-fold cross-validation, now used for live defect identification in production.
Used SHAP values and permutation importance to pinpoint which process variables actually caused manufacturing faults. These findings gave the engineering team concrete targets for improvement, measurably raising quality control benchmarks.
Customized Windchill PLM in Java with REST-based backend services, and automated the data pipeline between manufacturing sensors and dashboards, cutting manual data entry by over 60%.
Set up CI/CD with GitHub Actions to automate testing and deployment, cutting release cycles and reducing manual QA overhead.

Data Analyst Intern

Jun 2019 — Aug 2019

Leaf Innovations Pvt. Ltd., Delhi, India

Built a music recommendation engine using collaborative filtering and market basket analysis, achieving 82% accuracy in predicting user preferences across listening sessions.
Created Tableau dashboards that visualized user clusters and listening patterns, helping the product team make data-backed decisions about feature priorities.

Technical Arsenal

Technologies I work with

🧠

AI & Machine Learning

LLMsRAG AI AgentsLangChain LangGraphCrew AI Prompt EngineeringGuardrails Semantic CachingHybrid Search Fine-Tuned EmbeddingsPGVector NLP

⚡

Languages & Frameworks

PythonTypeScript SQLJava JavaScriptNext.js ReactFastAPI Drizzle ORMTailwind CSS

☁️

Cloud & DevOps

AWSGoogle Cloud Vertex AIDocker CI/CDGitHub Actions RedisMLOps SupabaseDrift Detection

📊

Data & Analytics

Random ForestXGBoost Scikit-learnPandas NumPyTableau Feature EngineeringData Wrangling PostgreSQLN8n

🛠️

Tools & Workflow

GitJira Agile/ScrumGemini API Claude APIGoogle Sheets API Windchill PLM

Projects

Selected work

Open Source GitHub

Agentic Content Pipeline

Multi-agent content system where three AI agents (Strategist, Writer, Reviewer) collaborate through a LangGraph stateful graph. Includes Redis semantic caching, automatic revision loops, human-in-the-loop approval, and SSE streaming.

LangGraphFastAPIRedisPydantic v2Docker

Open Source GitHub

RAG Hybrid Search Engine

Hybrid retrieval engine combining PGVector dense vectors with BM25 sparse search, fused via Reciprocal Rank Fusion. Features a dynamic query router that automatically picks the optimal retrieval strategy per query.

PGVectorBM25RRFFastAPILangChain

Platform

FIAE Learning Platform

Full-stack apprenticeship management system used daily across three vocational programs in Germany. Trainers manage curricula, trainees track their progress, and company admins oversee everything from a single dashboard.

Next.jsSupabaseDrizzle ORMDockerTypeScript

AI Assistant

HAI — Vocational AI Assistant

An AI assistant built specifically for vocational education. Routes simple lookups through SQL for instant, exact answers, and uses RAG with vector search for open-ended knowledge questions. Handles 90%+ of incoming support queries without human help.

PGVectorRAGHybrid SearchFastAPILLM

Automation

Autonomous Marketing Engine

Replaced the entire manual campaign planning process with an agentic pipeline. Automatically generates weekly marketing calendars with daily topics and posts them via Telegram. What used to take the marketing team a full week now takes hours.

LangGraphGemini 2.5 ProRedisFastAPI

ML Research

Fake News Detection (NLP)

NLP classification pipeline trained on a political dataset that identifies misinformation by analyzing linguistic patterns. Hit 92.82% accuracy with TF-IDF vectorization and passive aggressive classification, with feature importance analysis to make results interpretable and trustworthy.

NLPTF-IDFScikit-learnPython

Predictive Analytics

Lead Score Prediction

Predictive model for X Education that identifies which leads are most likely to convert. Used Recursive Feature Elimination and Variance Inflation Factor analysis to reach ~80% accuracy, sensitivity, and specificity. Found that time on site, lead source, and lead origin were the strongest conversion drivers, enabling targeted marketing strategies that improved engagement.

Logistic RegressionRFEVIF AnalysisPandasPython

Making AI work
where it matters most

Where I've delivered impact

Technologies I work with

Selected work

Open Source Contributions

Graphiti by Zep

Google ADK-Python by Google

Anthropic Courses by Anthropic

LangGraph by LangChain

Academic background

M.Sc in Data Science

Post Graduation in Data Science

B.Tech in Information Technology

Certifications & languages

Advanced Certification in Data Science

Data Science Methodology

Machine Learning with Python

Python for Data Science

Have a problem AI
could solve? Let's talk.

Making AI workwhere it matters most

Where I've delivered impact

Technologies I work with

Selected work

Open Source Contributions

Graphiti by Zep

Google ADK-Python by Google

Anthropic Courses by Anthropic

LangGraph by LangChain

Academic background

M.Sc in Data Science

Post Graduation in Data Science

B.Tech in Information Technology

Certifications & languages

Advanced Certification in Data Science

Data Science Methodology

Machine Learning with Python

Python for Data Science

Have a problem AIcould solve? Let's talk.

Making AI work
where it matters most

Have a problem AI
could solve? Let's talk.