AMEY NARWADKAR

AI / Machine Learning Engineer

“I build AI systems that move from theoretical research into resilient, high-throughput production reality.”

LOCATION Heidelberg, Germany

↓ SCROLL TO UNPACK NARRATIVE

I turn ideas into production.

From ambiguous ideas to reliable, production-ready AI systems.

01 · ARCHITECT

AI & LLM Systems

Turn ambiguous problems into agentic workflows, deterministic logic, and verifiable system specs.

Agentic Workflows Evals & Observability

LangGraph

Langfuse

Pydantic

OpenAI

Anthropic

02 · ENGINEER

Models, Retrieval & Reasoning

Fine-tuning, dense embeddings, knowledge graphs, and hybrid semantic re-ranking.

Knowledge Graphs & RAG Dense Retrieval & Reranking

PyTorch

Transformers

NetworkX

FlashRank

ONNX

03 · SCALE

Services & Data

High-throughput async APIs, vector/relational datastores, and latency/cost optimization.

Async APIs & Microservices Vector & Relational Stores

Python

TypeScript

FastAPI

PostgreSQL

Redis

04 · DEPLOY

Infrastructure & Delivery

Containerization, CI/CD automation, production telemetry, and feedback loops.

Docker & Cloud Infra CI/CD & Telemetry

Docker

Linux

Git & GitHub

CI/CD

AWS

Selected Work

Production platforms & verified research architectures.

VISIT TRACXO.DE ↗

01 / 03

TRACXO

Conversational multi-agent platform that parses unstructured vendor bills and POS data to track ingredient costs, detect price leakage, and deliver instant procurement answers on WhatsApp.

50+ Live deployments

< 850ms P95 tool latency

RESEARCH ARCHITECTURE

02 / 03 AGENTIC RAG

Self-Corrective Retrieval Engine

Deterministic 9-node state graph with query decomposition, hybrid BM25 + pgvector retrieval, and automated critique gates.

0.0% False Citations

9 Nodes State Machine

MASTER'S THESIS // HEIDELBERG UNIVERSITY

03 / 03 KNOWLEDGE GRAPHS

Rule-Guided Graph Reasoning

Leakage-free link prediction marrying RotatE complex embeddings with AnyBURL first-order Horn rules across 14.8M+ triples.

14.8M+ KG Triples

100% Rule Explainability

Experience

NEC Labs Europe

Working Student — Software Development

May 2025 — Present · 1 yr 6 mos Heidelberg, Germany

Built literature-grounded synthetic data generation pipelines (375+ published correlations, 11+ evidence packages) using multi-agent planning, generation, critique, and verification to produce realistic longitudinal data.
Reduced synthetic data quality loss by 81% (0.106 to 0.020) on a 230+ user data benchmark, expanding validation to 6 quality gates and eliminating constraint violations entirely.
Outperformed statistical baselines with 2.4× lower ICC error and 26% lower holdout RMSE across 5,000+ longitudinal gap-filling predictions.
Designed schema-constrained LLM pipelines with automated validation/repair, achieving 100% structured-output parse success while tracking token usage and cost.

✦ MOMENTS AT NEC LABS

Soul AI

AI Data Specialist — LLM Evaluation

Oct 2023 — Apr 2024 · 7 mos Remote

Contributed to RLHF data pipelines evaluating LLM outputs for college-level mathematics — multi-turn reasoning, hallucination reduction, and context-retention analysis.
Reviewed and annotated model outputs across 500+ evaluation tasks, identifying failure patterns and collaborating with human evaluators to improve model quality.

Education

Grounded in scientific computing, mathematical rigor, and neuro-symbolic machine learning.

Heidelberg, Germany

Heidelberg University

M.Sc. Scientific Computing Hans-Peter Wild Talent Scholarship

Advanced coursework in numerical methods, machine learning, and high-performance computing.

Scientific ML Numerical Methods Optimization

Pune, India

Fergusson College

Bachelors in Mathematics CGPA 9.63 / 10.0 · Top 5 in Class

Foundations in linear algebra, real analysis, differential equations, and computational mathematics.

Linear Algebra Real Analysis Differential Equations

Deep Cosmic Space with Glowing Planet Horizon

Let's
build what's
next.

IDEAS → RESEARCH → REAL-WORLD IMPACT

GET IN TOUCH

ameynarwadkar1729@gmail.com

COPIED TO CLIPBOARD ✓

Shoot me an email. I'd love to hear from you.