AuraMind
Deutsch
Try Voice ↗
Kartik BK Mishra

Founder portfolio

Kartik BK Mishra

Founder of AuraMind. AI automation, agent workflows and software.

Hamburg, Germany

public repositories on GitHub
70counted 11 October 2026
AuraMind founded
2026February 2026, Hamburg
Degree completed in Hamburg, Germany
Master’s
Languages
3English, Hindi, German

Selected projects

What I have built. How it works.

Client work, working automations and independent engineering. Each card shows the problem, what I built, the result and its limit.

A server terminal starting a local language model, scene from the company film
Still from the AuraMind company film. Illustration.

Client work · Local AI

Local agents on a Mac mini

Problem
The client wanted agents running on their own Mac mini and needed to know what 16 GB of memory could handle.
What I built
Installed and debugged OpenClaw and Ollama, tested the limits of a 16 GB Mac mini, and delivered a setup script and client guide.
Result
An evaluated setup, installation script and practical operating guide.
Limit
Memory set the limit. For some workloads a cloud model was more practical.
  1. Workload
  2. Hardware check
  3. Agent setup
  4. Handover
  • OpenClaw
  • Ollama
  • Claude Code
  • macOS
Watch a local model workflow (sample data) Discuss a similar project
Evidence and scope
The input
A Mac mini with 16 GB memory and a business automation brief.
The output
An evaluated setup, installation script and practical operating guide.

What the evidence shows

The evaluation documented what worked, memory constraints and where a cloud model was more practical. This installation is separate from the local model training service now offered by AuraMind.

A phone call in progress next to a calendar, scene from the company film
Still from the AuraMind company film. Illustration.

Client work · Voice

Voice reception for garages

Problem
A garage wanted calls answered in French, with the vehicle, contact details and preferred appointment time captured.
What I built
Configured French reception instructions for answering questions and collecting vehicle details, contact information and preferred appointment times.
Result
A configured reception assistant with instructions for appointment-request intake.
Limit
The evidence is a draft configuration. No call logs, completed bookings or calendar connection are shown.
  1. Caller question
  2. Intake instructions
  3. Details to confirm
  4. Request recap
  • Vapi
  • Conversation design
  • Input validation
Watch a workshop call (sample data) Discuss a similar project
Evidence and scope
The input
Garage reception requirements for caller details and appointment preferences.
The output
A configured reception assistant with instructions for appointment-request intake.

What the evidence shows

The supplied screenshot shows a draft assistant configuration and French reception instructions. Call logs, completed bookings and calendar or CRM connections are not included in that evidence.

A baker photographing a tart, with a social post waiting for approval, scene from the company film
Still from the AuraMind company film. Illustration.

Client work · Content

Research and content operations

Problem
An existing OpenClaw deployment needed competitor research, German reports and social content, and some runs were failing.
What I built
Extended the deployment for research, reports, social content and newsletters. Investigated failed runs and checked delivery before retrying.
Result
German reports and content, with delivery checks and a visible run history.
Limit
The original deployment existed before this work. Published posts are recorded separately.
  1. Source watchlist
  2. Research
  3. Review
  4. Delivery check
  • OpenClaw
  • Google Workspace
  • SocialBu
  • MCP
Watch a content workflow (sample data) Discuss a similar project
Evidence and scope
The input
Research sources, scheduled jobs and existing business integrations.
The output
German reports and content, with delivery checks and visible run history.

What the evidence shows

Project records include completed research and delivery checks, with published and scheduled posts recorded separately. The original deployment existed before this engagement.

An n8n workflow with connected steps, scene from the company film
Still from the AuraMind company film. Illustration.

Own build · Research

Research with traceable sources

Problem
Research drafts needed sources a person could check before anything was saved.
What I built
Connected n8n and Hermes to gather public sources, prepare German drafts and check the returned fields and evidence before saving the work for review.
Result
Source-linked research and German drafts saved for review.
Limit
Manual research runs only. No automated outreach or social publishing.
  1. Public sources
  2. Hermes task
  3. Validation
  4. Saved draft
  • n8n
  • Hermes
  • Grok
  • REST APIs
Watch a research workflow (sample data) Discuss a similar project
Evidence and scope
The input
Public company sources and a structured research task.
The output
Source-linked research and German drafts saved for review.

What the evidence shows

The records show successful manual research runs and a corrected validation retry. This workflow does not demonstrate automated outreach or social publishing.

An invoice checked against an order record, scene from the company film
Still from the AuraMind company film. Illustration.

Own project · AI evaluation

Test the answer. Test the grader.

Problem
An AI answer can look right and still be wrong, and a weak grader can reward it.
What I built
Built a document reconciliation task with fictional business data, a reference solution, 12 scoring criteria and a separate containerized verifier.
Result
A defensible reference answer and a verifier tested against incorrect and empty submissions.
Limit
An authored evaluation package, not a client benchmark or a model ranking.
  1. Conflicting records
  2. Reference answer
  3. Grading checks
  4. Verified result
  • Python
  • SQL
  • Docker
  • PDF · XLSX · JSON
Watch an evaluation workflow (sample data) Discuss a similar project
Evidence and scope
The input
Conflicting fictional PDFs, spreadsheets, JSON and a bank ledger.
The output
A defensible reference answer and a verifier tested against incorrect submissions.

What the evidence shows

Recorded validation on 7 October 2026 shows 68 passing local tests and Docker checks for correct and empty submissions. It is an authored evaluation package, not an accepted client benchmark or a measured model ranking.

Open Harness prototype screen where two models agree on a checked equation
Actual prototype screen

Personal project · In development

Open Harness

A desktop workspace for coordinating Claude Code and Codex, with context handoffs, approval handling and separate verification of agent work.

  • TypeScript
  • Agent adapters
  • Context management
  • Verification
Inspect Open Harness

Skills, shown by the work

Each skill links to the project that shows it.

Experience and qualifications

Roles since 2020, newest first. Education, certificate and languages follow.

  1. March 2026 to presentCurrent

    Freelance AI tool-use trainer and model evaluator at Independent, contract

    Prompts, task scenarios and rubrics for multimodal and tool-use AI training. Reviews outputs for accuracy, workflow order and hallucinated tool behaviour. Documents failure patterns for frontier AI clients.

  2. February 2026 to presentCurrent

    Founder at AuraMind

    Maps manual workflows and builds practical AI automation: email, invoicing, lead capture, CRM updates and reporting with n8n, Make, Zapier, Vapi and Claude-based workflows. Delivered AI tool training to schools and local authorities in Germany.

  3. October 2025 to January 2026

    AI red teamer and LLM evaluator at Mercor

    Adversarial testing of LLMs in English and Hindi. Designed multi-cycle QA frameworks with rubrics and prompt skeletons across seven evaluation cycles.

  4. March 2025 to July 2025

    Automation lead, project based at BackwardsLA

    AI-driven LinkedIn outreach automation and tracking dashboards.

  5. October 2024 to January 2025

    Process optimisation intern at Adlevate

    Data analysis, automation bots and process-optimisation feasibility work.

  6. October 2022 to December 2022

    Fulfilment associate at Amazon Logistik und Operations

    Work-study role at the Winsen fulfilment centre.

  7. February 2021 to August 2021

    Front-end developer intern at Full Stack Learning

    Front-end development internship, Jaipur.

  8. January 2020 to December 2020

    QA engineer at Shriwari Enterprises

    Quality assurance engineering role, India.

Education and certificates

  • Bachelor’s in engineering, Master’s degree from Hamburg, Germany
  • Data Analytics with GenAI, upGrad and Adda247, February 2026

Languages

  • English, native
  • Hindi, native
  • German, conversational

Work with me

Send the task, the outcome you want and the one action that must never go wrong.

Start a projectBack to AuraMind