Skip to content
View Akshat4112's full-sized avatar
🏠
Working from Home
🏠
Working from Home

Organizations

@Robofied

Block or report Akshat4112

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
akshat4112/README.md

Akshat Gupta

Applied AI engineer and researcher

I build production AI systems at additiv, working from Stuttgart, Germany. My current focus is agentic workflows and document intelligence for finance and insurance: grounding outputs in evidence, validating structured data, and evaluating complete systems.

My earlier work spans NLP, speech, computer vision, and adversarial ML. I care about making AI systems observable, testable, and useful in real workflows.

Portfolio · Writing · Publications · LinkedIn · Kaggle

Current focus

  • Reliable agentic and retrieval-augmented AI systems
  • Evaluation of nondeterministic LLM workflows
  • Document intelligence and knowledge representation
  • AI security, privacy, and auditability
  • Speech processing and speaker anonymisation

Research and engineering projects

A public browser application with 14 tools for reading, organising, and converting PDFs. Selected documents are processed on the device; loading the site and its assets still connects to GitHub Pages.

Engineering focus: independent validation of downloaded outputs, browser resource cleanup, and explicit limits on supported document features.

Open application · Code · Engineering case study

A visual approach to detecting homoglyph phishing domains. Domain names are rendered as images and classified with an attention-based convolutional network.

My contribution: research, dataset construction, model development, and evaluation.

Project page · Paper · Code · Dataset

An experimental diffusion model for generating speaker embeddings, developed to study speaker anonymisation and the privacy–utility trade-off in speech systems.

Focus: diffusion modelling, speaker representations, anonymisation, and speech synthesis experiments.

Code and experiments · Technical article

A collection of reproducible notebooks covering classical machine learning and deep learning across NLP, computer vision, forecasting, and tabular problems.

Explore the notebooks

In development

  • ImmoPilot DE — unreleased: German property-purchase, financing, and scenario comparison. Version 1 release-readiness work is in progress.
  • NettoPilot DE — in development: a German salary calculator and job-offer comparison tool, with application foundations and calculation work in progress.

Research notes and learning resources

  • Research paper presentations — presentations from my MSc in Computational Linguistics, covering speech, NLP, psycholinguistics, parsing, and machine translation.
  • Awesome NLP Resources — a curated collection of foundational and modern NLP material.

Open datasets

  • FinPlainLM — financial plain-language preference data with supervised fine-tuning and direct preference optimisation splits.
  • GlyphNet Homoglyph Domains — paired domains and rendered images for homoglyph-phishing research.

Technical writing

I write technical notes and case studies about document software and production AI systems, including:

Browse all articles on my portfolio.

Current stack

Python · Azure OpenAI · Agno · Langfuse · Pydantic · uv · Docker

Working areas

LLM systems · RAG · AI agents · Evaluation · Document intelligence · PyTorch · TensorFlow · NLP · Speech processing · Knowledge graphs · MLOps · AWS

Contact

For research, engineering, or collaboration conversations:

Pinned Loading

  1. Awesome-NLP-Resources Awesome-NLP-Resources Public

    This repository contains landmark research papers in Natural Language Processing that came out in this century.

    212 54

  2. school-erp school-erp Public

    SchoolERP Project is School ERP System which has various features as students management,accounts management,attendance management,faculty management,HR management,other expenses management

    PHP 319 193

  3. Glyphnet Glyphnet Public

    Detecting homoglyph (IDN homograph) phishing domains by rendering domain names as images and classifying them with an attention-based CNN. Includes a 2.5M-image real-vs-homoglyph dataset (on Huggin…

    Python 6

  4. akshat4112.github.io akshat4112.github.io Public

    personal website

    HTML