Tarun Gupta — AI Systems Architect · LLM & MLOps Consultant

Open Source & Personal Projects

Projects

⭐ Featured

MindMorph — Adaptive AI Learning Platform

Multi-agent LangGraph platform with adaptive mastery tracking. Groq-powered LLaMA 3.3-70B Orchestrator routes to 5 graphs: SCOUT (Academic + Market + Practical + Factual → Consensus → Synthesizer → Reviewer → Mermaid Skill Graph), CONTENT (Boost / Builder / Sprint), LEARNING PLAN (mastery-gated node progression, ≥80% mastered), LESSON (structured delivery + DuckDuckGo + job-scraper research), EXERCISE (coding challenges + live unit-test grader via streamlit-ace). Full LangSmith tracing. GitHub MCP.

0Agents

0LangGraph Workflows

0Content Formats

0Python

LangGraphLangChainGroq / LLaMA 3.3 StreamlitLangSmithMCP Streamlit-AcePython

⎋ GitHub → Case Study

⭐ Featured

AI Content Machine

Autonomous content studio for 3 niches (Data Science · Life · Poetry). 150 Python scripts on the Claude CLI with per-task model routing (Fable 5 · Sonnet 5 · Opus 4.8 · Haiku): idea scoring → one-run blog pipeline (posts, carousels, decks, worksheets, reel briefs) → V2 video pipeline (storyboard → HyperFrames → FFmpeg, word-synced captions) → APScheduler daemon auto-publishing from SQLite across 7 platforms. 32-file virality knowledge base + closed-loop analytics with UTM star attribution.

0Pieces/Month

0Python Scripts

0Platforms

0YouTube Channels

PythonClaude CLI · Model RoutingHyperFrames FFmpegAPSchedulerMeta Graph API SQLitelaunchd

⎋ GitHub → Case Study

⭐ Featured

AI Job Engine — Job Search Automation

12 portal scrapers (LinkedIn · Greenhouse · Lever · Indeed · Naukri · Wellfound · Instahyre · ArbeitNow · Hirist · Remotive · WeWorkRemotely · Hacker News). Gemini Flash fit scoring. Dynamic resume tailoring per JD. Playwright one-click Easy Apply. Streamlit dashboard.

0Portal Scrapers

0Jobs Tracked

0Automated

0Faster Hunt

PythonGemini FlashPlaywright BeautifulSoupStreamlitSQLite

⎋ GitHub → Case Study

⭐ Featured

Autopilot Job Hunt

Open-source AI job agent — pip install autopilot-jobhunt. Scans 130+ company careers pages nightly, scores every role 0–100 against your resume with an LLM, pushes top matches to Telegram, and drafts a tailored resume + cover letter on demand in 60s. Claude Code MCP mode lets you run the whole pipeline in natural language.

0GitHub Stars

0Forks

0Careers Pages

$0LLM API Cost

PythonClaude MCPOpenRouter TinyFish APITelegramPyPI

⎋ GitHub → Case Study 📦 PyPI

⭐ Featured · New

Inbox to Action — Agentic Email Triage

One command triages your inbox in a single agentic pass: classify (action / FYI / newsletter / noise), summarize threads, extract tasks, draft replies — never sends (readonly + compose scopes only, enforced by test). BYOK across 6 LLM providers including fully-local Ollama and keyless Claude Code. pip install inbox-to-action.

0LLM Providers · BYOK

0Tools → 1 Pass

0Emails Auto-Sent

100%Local Option (Ollama)

PythonMCP ServerGmail API OllamaOpenRouterDocker / GHCR PyPI

⎋ GitHub 📦 PyPI → Case Study

⭐ Featured · New

Fable Mode — Fable-Class Behavior for Any Claude Model

Open-source Claude Code plugin. A SessionStart + SubagentStart hook injects an always-on ruleset into every session and subagent, distilling Fable 5's discipline into Opus, Sonnet, and Haiku: outcome-first answers, whole-request coverage, act-don't-ask execution, minimal scope, honest verification. Five reliable failure modes → five rules, plus §6 — a stronger-model consult on Haiku/Sonnet. /plugin marketplace add tarunlnmiit/fable-mode.

0Failure Modes Fixed

0Rule Sections

0GitHub Stars

0Models Covered

Claude CodePluginSessionStart Hook MarketplaceJavaScriptMIT

⎋ GitHub → Case Study

⭐ Featured

AI CV Engine — JD-Aware CV & Cover Letter Generator

YAML-driven CV tailoring engine. NLP keyword scoring ranks projects/skills by JD match. 12 role-profile classifiers (LLM/GenAI/Agent/FDE/DS/MLOps/…) each with distinct title, summary opener, and cover letter. python-docx builder → DOCX/PDF. Batch mode from Excel job tracker. 30+ pytest assertions.

0Role Classifiers

0Pytest Assertions

0Automated

0Manual Editing

PythonClaude AIYAML python-docxLibreOffice

⎋ GitHub → Case Study

Ship it like a product

Open Source & Distribution

Side projects that are installed, not just starred — published to PyPI, shipped as Claude Code plugins, wired into the MCP ecosystem (Official MCP Registry · Glama · Smithery manifests), shipped as Docker images on GHCR, CI-tested on every push, MIT-licensed. Bring your own key: every tool runs against your choice of LLM provider — including fully local.

autopilot-jobhunt

★ 164 · 43 forks

pip install autopilot-jobhunt

Autonomous job-search agent: scans 130+ careers pages nightly, LLM-scores every role 0–100 against your full resume, alerts via Telegram, drafts tailored resume + cover letter on demand. 4-model OpenRouter fallback chain → $0 API cost. Runs 24/7 on a Raspberry Pi.

PyPIMCP · listed on GlamaDockerGitHub Actions CIMIT

⎋ GitHub 📦 PyPI → Case Study

inbox-to-action

BYOK · 6 providers

pip install inbox-to-action

One-pass agentic inbox triage: classify, summarize, extract tasks, draft replies (drafts only — no send scope exists in the codebase, enforced by test). Provider-agnostic BYOK: Ollama (fully local), keyless Claude Code, OpenRouter, OpenAI, NVIDIA NIM, Anthropic. MCP server + Claude Code skill included.

PyPIMCP Registry · Smithery · GlamaDocker / GHCRCIMIT

⎋ GitHub 📦 PyPI → Case Study

fable-mode

★ 0 · MIT

/plugin marketplace add tarunlnmiit/fable-mode

Fable-class discipline on any Claude model. A SessionStart + SubagentStart hook injects an always-on ruleset into every session and subagent — five failure modes → five rules, plus a stronger-model consult on Haiku/Sonnet. Zero config, on by default.

Claude Code PluginMarketplaceSessionStart HookMIT

⎋ GitHub → Case Study

164+

GitHub Stars Across Tools

Packages on PyPI

MCP Servers Shipped

LLM Providers · BYOK

24/7

Self-Hosted on Raspberry Pi

Claude Code Plugin

Work History

Experience

Generative AI & LLM Systems

Voice AI Agents for Customer Service

May 2025 – Jan 2026

LPDG Group · Managing Consultant — AI & Data Science · Retail/E-commerce · Germany 🇩🇪

Deployed end-to-end voice AI agents (STT → LLM → TTS) handling inbound customer service calls autonomously — integrated with order management + inventory via tool calls for full resolution without human escalation
~65% calls auto-resolved · multi-FTE workload absorbed · ~40% average handle-time reduction
Stack: CrewAI · VAPI · n8n · Twilio · Azure OpenAI · Azure STT/TTS · Streamlit

Intelligent Invoice Entity Recognition Platform — $100K+/month savings

Apr 2023 – Present

LPDG Group · Managing Consultant — AI & Data Science · Utilities · India 🇮🇳

FTP → Azure Document Intelligence → LLM → Snowflake → Streamlit pipeline processing multi-format PDFs (1–1,000 pages); confidence-gated human review prevented silent errors
Service-address normalisation against ground-truth source; Snowflake-native with zero external orchestration
Stack: Python · Snowflake · Azure Document Intelligence · Azure OpenAI · Streamlit · GitHub Actions CI/CD

7 Production RAG Chatbots — 1 Plug-and-Play Framework

Apr 2023 – Present

LPDG Group · Managing Consultant — AI & Data Science · Multi-client 🌐

Shipped 7 production chatbots from one client-agnostic framework — new client = new config, not new codebase; multi-turn conversation via LangChain Retrieval Agents + Conversation Memory
FAISS + Azure AI Search for semantic retrieval; FastAPI async; Kubernetes auto-scaling + Terraform-provisioned infra
Led cross-functional team of 4 engineers + 3 client SMEs across all 7 deployments

LLM Product Intelligence Dashboard

May 2022 – Feb 2023

Vivere GmbH · Applied Data Scientist · Hamburg, Germany 🇩🇪 · 9-month fixed-term

Built LLM-powered product intelligence dashboard (Azure OpenAI + BigQuery + PostgreSQL) — non-technical product teams query market data conversationally
Competitor clustering, organic-vs-inorganic spend attribution; event-driven Asana/email alerting via FastAPI/Pydantic
Self-serve Streamlit UI eliminated manual competitor research across product and marketing teams

MLOps & Forecasting Pipelines

Reusable MLOps Forecasting Framework — 5 Use Cases, 1 Codebase

May 2025 – Jan 2026

LPDG Group · Managing Consultant — AI & Data Science · Manufacturing & E-mobility · Germany 🇩🇪

One parameterised pipeline (pre-process → train → infer → post-process) serving 5 assembly-line + e-mobility forecasting use cases with zero code duplication
Eliminated costly Snowflake → Databricks → Snowflake round-trips; resolved missing sensor time-series via behavioural clustering — enabling reliable forecasting where previously impossible
Snowflake Model Registry + Feature Store; GitHub Actions CI/CD across all 5 pipelines

Large-Scale Text Processing & Email Generation

Apr 2023 – Mar 2024

LPDG Group · Managing Consultant — AI & Data Science · India 🇮🇳

GPT-3.5 + Apache Spark on AWS EMR for large-scale unstructured text → structured email conversions
Airflow for scheduling, dependency management, and failure recovery across batch jobs
Processed 100K+ documents; maintained Snowflake fact table for downstream analytics

Computer Vision & Anomaly Detection

YOLOv11 Radiator Classification (Edge Deployment)

Feb 2026 – Present

LPDG Group · Senior AI/ML Engineer · India 🇮🇳

YOLOv11 ensemble (3 models) handles class imbalance across radiator types — classifies radiator serviceability from single field image with 95%+ accuracy
Browser-based WebAssembly inference — zero cloud calls, fully offline in remote field conditions with no connectivity
Deployed on Snowflake Container Services; back-office Streamlit app enables continuous retraining without engineering intervention

Real-Time Gesture & Shape Tracking

Apr 2024 – Apr 2025

LPDG Group · Senior AI/ML Engineer · India 🇮🇳

Real-time human gesture tracking via OpenCV + pose estimation — live demonstration at WETEX Dubai (major Middle East energy exhibition)
Sub-100ms latency on standard hardware; Streamlit demo interface for stakeholder engagement

ML Algorithm Benchmarking & Anomaly Detection

Apr 2024 – Apr 2025

LPDG Group · Senior AI/ML Engineer · India 🇮🇳

Benchmarked 10+ ML algorithms against 15+ class imbalance strategies (SMOTE, class weights, threshold tuning, ensemble methods)
Automated end-to-end anomaly detection pipeline on AWS (S3, Lambda, SageMaker, Docker, Jenkins)
Selected optimal config via grid search; deployed monitoring dashboard for real-time anomaly alerts

Research & Explainable AI

Data Scientist (Research) — AI & XAI

Oct 2020 – Mar 2022

IBMI, Universitätsklinikum Magdeburg · Magdeburg, Germany 🇩🇪

Active learning on BraTS2017 medical imaging: Query-By-Committee (KLD · JSD · Entropy) — 7% improvement in informative sample selection over random; precision fluctuation reduced 12% → 1%, cutting annotation cost
Explainable Deep RL: Dropout DQN + saliency maps — 3–6% performance improvement using only 2.5–10% training data via MMD-Critic selection; IRTEX image retrieval with 80% majority user approval (85% global, 65% local explanations)
MSc thesis: Human Brain Simulation (VisMem cognitive model on Nengo) · IEEE paper: IRTEX — Image Retrieval with Textual Explanations (XAI), ieeexplore.ieee.org/document/9582666

Backend & Full-Stack Infrastructure

Automated MySQL schema migration framework supporting diverse legacy schemas — eliminated manual migration risk
Built 200+ test end-to-end framework achieving full server-side coverage; reduced production bugs from migrations
Validated schema consistency, data integrity, and performance across migrations

Database Migration & Testing Framework

Apr 2019 – Sep 2020

3DQR GmbH · Software Engineer — Data & Backend · Magdeburg, Germany 🇩🇪

Automated MySQL schema migration framework supporting diverse configurations — eliminated manual migration risk
200+ end-to-end test suite (Mocha/Chai) covering all server-side paths — enabled confident ship cadence without manual regression

Software Engineer — Data & Backend

Jan 2016 – Oct 2018

Careers360 · ConveGenius Group · Delhi / Noida, India 🇮🇳

LMS delivery managing team of 3 (Django · Node.js · MySQL · AWS); built 24×7 real-time student-query chat system
Secure Android REST APIs with server-side business logic; analytics dashboarding for student-engagement and platform-health metrics

Peer-reviewed & presented

Research & Publications

IEEE Paper

IRTEX — Image Retrieval with Textual Explanations (XAI)

Explainable image retrieval with heatmaps + saliency maps at set and individual level. 80% majority user approval (85% global, 65% local explanations), validated via user studies. Published at IEEE — ieeexplore.ieee.org/document/9582666.

Research

Active Learning on Medical Imaging (BraTS2017) — Universitätsklinikum Magdeburg

Query-By-Committee (KLD · JSD · Entropy): 7% improvement in informative sample selection over random; precision fluctuation reduced 12% → 1% — cutting medical-image annotation cost.

Research

Explainable Deep Reinforcement Learning — Data Efficiency over Data Volume

Dropout DQN + saliency maps on MNIST / Space Invaders: 3–6% performance improvement using only 2.5–10% of training data via MMD-Critic sample selection.

MSc Thesis

Human Brain Simulation — VisMem Cognitive Model on Nengo

MSc Digital Engineering thesis, Otto-von-Guericke-Universität Magdeburg — computational cognitive modelling of visual memory.

Live Demo

Real-Time Gesture AI — WETEX Dubai

Live demonstration of real-time human gesture tracking (OpenCV + pose estimation, sub-100ms latency) at WETEX Dubai, a major Middle East energy exhibition.

I turn scattered AI capabilities into tools people can actually run.

What I bring

Skills

🤖 AI / LLM

⚙️ MLOps & Infra

👁️ Computer Vision

☁️ Cloud & Backend

🐍 Languages & Frameworks

📊 ML / Data Science

Projects

MindMorph — Adaptive AI Learning Platform

AI Content Machine

AI Job Engine — Job Search Automation

Autopilot Job Hunt

Inbox to Action — Agentic Email Triage

Fable Mode — Fable-Class Behavior for Any Claude Model

AI CV Engine — JD-Aware CV & Cover Letter Generator

Open Source & Distribution

autopilot-jobhunt

inbox-to-action

fable-mode

Experience

Generative AI & LLM Systems

Voice AI Agents for Customer Service

Intelligent Invoice Entity Recognition Platform — $100K+/month savings

7 Production RAG Chatbots — 1 Plug-and-Play Framework

LLM Product Intelligence Dashboard

MLOps & Forecasting Pipelines

Reusable MLOps Forecasting Framework — 5 Use Cases, 1 Codebase

Large-Scale Text Processing & Email Generation

Computer Vision & Anomaly Detection

YOLOv11 Radiator Classification (Edge Deployment)

Real-Time Gesture & Shape Tracking

ML Algorithm Benchmarking & Anomaly Detection

Research & Explainable AI

Data Scientist (Research) — AI & XAI

Backend & Full-Stack Infrastructure

Database Migration & Testing Framework

Software Engineer — Data & Backend

Certifications & Education

Certifications

SnowPro Core

SnowPro Specialty: Gen AI

Introduction to Agent Skills

Building with the Claude API

Introduction to MCP

Claude Code in Action

AWS Machine Learning Specialty

Education

MSc Digital Engineering

BTech Computer Science

Research & Publications

IRTEX — Image Retrieval with Textual Explanations (XAI)

Active Learning on Medical Imaging (BraTS2017) — Universitätsklinikum Magdeburg

Explainable Deep Reinforcement Learning — Data Efficiency over Data Volume

Human Brain Simulation — VisMem Cognitive Model on Nengo

Real-Time Gesture AI — WETEX Dubai

Writing & Content

You're Still Paying $200/Month for AI Tools You Could Replace With a Free Local Setup Tonight

Python Silently Lied to You (And You Didn't Notice)

I Stopped Googling Travel Guides. I Interrogated an AI Instead — the Exact Prompt That Worked

What actually ships — a note every few weeks

Contact

Email

Phone

LinkedIn

GitHub

Medium

YouTube

Send me a message