Sandipan Seal sandipanseal

🧠 About Me

role:      AI Research Engineer @ aiio GmbH
based_in:  Magdeburg, Germany 🇩🇪
origin:    Kolkata, India 🇮🇳

mission: >
  Take frontier AI out of the notebook and into production:
  cost-aware, observable, safe, and genuinely useful to users.

what_i_do:
  - 🏗️  Build complete LLM products for clients, from requirements to deployment
  - 🕸️  RAG-enabled multi-agent systems, knowledge graphs and chatbots over process data
  - ✨  AI suggestions, recommendations and AI-generated descriptions inside aiio's products
  - 🤝  Work directly with clients as a forward-deployed engineer

focus_areas:
  - Multi-agent orchestration · RAG · Knowledge graphs (FalkorDB, Cypher, Neo4j)
  - LLM gateways · cost-aware routing · guardrails · evals (LLM-as-judge, RAGAS)
  - LLM security · prompt-injection defense · PII redaction
  - MCP · Claude Agent SDK · DSPy · open-weights serving (Ollama, vLLM)
  - AWS serverless · Terraform · Docker · Kubernetes · OpenTelemetry

research:
  master_thesis: "From RF Embeddings to Semantic Reasoning: Explainable Jamming
                  Analysis via LLM-Based Evidence Interpretation" (OVGU Magdeburg)

highlights:
  - 📜 Indian Patent IN 202031037602 A, Automated Bag Valve Mask
  - 🏆 Winner, Conquer COVID-19 Hackathon
  - 📄 2x Best Paper Award (I3SET 2k19, NCMR 2020)
  - 🌟 2x High Five Team Award @ Persistent Systems

🤖 GenAI & Agents
🧰 LLM Frameworks & Providers
🧪 ML & Data Science
💻 Languages
⚙️ Backend & Frontend
🗄️ Databases & Graphs
☁️ Cloud, DevOps & Observability
🔒 Testing & Security

⚡	AI Research Engineer — aiio GmbH _{Oct 2026 – Present · Magdeburg, Germany} Gave clients 70–80% more clean, usable data by building and deploying end-to-end RAG-enabled multi-agent systems that ingest process data, extract workflows, and use LLMs to build knowledge graphs and chatbots. Cut clients' consulting spend by researching and shipping software products that act as an automated consultant. Turned ambiguous problems into working solutions as a forward-deployed engineer: requirements with clients, then design, build, and deployment.
🌱	Working Student — AI/ML Software Development — aiio GmbH _{Jan 2025 – Sep 2026 · Magdeburg, Germany} Added AI suggestions, recommendations, and AI-generated descriptions to aiio's main process-data product and Polaris by building LLM-based features into the live products. Accelerated delivery of customer-facing AI tools by building responsive React.js / Next.js front-ends integrated with production models.
🧠	Research Assistant (Hiwi) — Leibniz Institute for Neurobiology _{Mar 2025 – Jun 2026 · Magdeburg, Germany} EEG signal analysis and depth-perception research · `Python`, `OpenGL`, `MATLAB`, `Pygame`, `Signal Processing`.
🚀	Senior Software Engineer — Persistent Systems _{Sep 2021 – Mar 2024 · Kolkata, India} Raised healthcare-diagnostic prediction accuracy by ~20% for Adaptive Biotechnologies by deploying predictive ML models into production. Delivered enterprise features end to end for PerkinElmer across React, .NET / C# and MongoDB. Cut regression-test cycle time by ~30% with automated QA suites (Selenium, API testing). High Five Team Award twice (Jun & Sep 2023).
🔧	Technical Intern — West Bengal Electronics Industry Dev. Corp. (Webel) _{Jan 2020 – Mar 2020 · Kolkata, India} Designed and built the Automated Solar Panel Street Light: circuit tracing, soldering, power control unit testing.

🎓 Master's Thesis — Explainable RF Jamming Analysis with an LLM Evidence Interpreter From RF Embeddings to Semantic Reasoning: Explainable Jamming Analysis via LLM-Based Evidence Interpretation · OVGU Magdeburg Experimental study of how LLMs (as evidence interpreters) and conventional ML (as classifiers) perform on synthetic RF-jamming data. Five LLMs, one identical protocol. A supervised four-axis LDA evidence lifted regime accuracy from chance (19% → 56%, 73% with a deterministic correction rule), citations were 100% grounded but not always causally faithful (paired perturbation tests, McNemar/Holm, group-aware bootstrap), and the same four LDA features without any LLM reach 80–81%: the representation, not the model, does the work. `LLMs` `Explainable AI` `LDA` `Statistics` `Python`
📊 DeliveryLens-AI AWS-native software-delivery intelligence platform connecting Jira work items with GitLab merge requests and CI pipelines. Calculates auditable flow metrics, diagnoses bottlenecks, recommends approval-gated interventions with ROI estimates, and verifies throughput improvements. Guarded Text2SQL, Glue + Athena analytics, offline DSPy optimization, Terraform serverless deployment. `FastAPI` `Jira REST` `GitLab GraphQL` `DSPy` `PySpark` `AWS Glue` `Athena` `Step Functions` `Terraform`
🛡️ AegisOps-AI Agentic AI Incident Commander. Evidence collection (logs, metrics, k8s, deploys) → LLM root-cause analysis → safety classification → approval-gated runbooks. Pluggable LLM backend, end-to-end Prometheus + Grafana + OpenTelemetry. `FastAPI` `Next.js` `Postgres` `Prometheus` `Grafana` `OpenTelemetry`	🚦 InferOps-AI Production-grade LLM gateway: cost-aware routing, PII redaction, prompt-injection blocking, RAG, Redis caching, evals (LLM-as-judge + RAGAS), full observability. `FastAPI` `LangChain` `Qdrant` `Redis` `Prometheus` `Grafana`
⚖️ AI-Compliance-Gatekeeper AI compliance & governance system that analyzes policies, documents and workflows to detect regulatory risks and prevent data leakage. `n8n` `FastAPI` `RegTech` `Python`	🧪 Multi-Agent QA — CrewAI Multi-agent system (explorer, scenario writer, script generator, auto-repairing runner) turning natural-language requirements into executable Playwright tests. `CrewAI` `OpenAI` `FastAPI` `Playwright` `React`
🏛️ Mine@TENJI — Legal Search Speech-recommendation system for parliamentary debates and legislative records. `SentenceTransformers` `FAISS` `NLP` `IR` `TypeScript`	🌐 Company Website Clustering Unsupervised clustering of company websites for market segmentation. `Python` `NLP` `Clustering`
🔬 Embedded HW + Cone Detection Hardware evaluation pipeline with OpenCL, OpenCV and FFT-based cone detection. `OpenCL` `OpenCV` `FFT` `Embedded`	🫁 Automated Bag Valve Mask — 📜 Patent IoT-based automated BVM for COVID patients. Indian Patent IN 202031037602 A · 🏆 Winner, Conquer COVID-19 Hackathon. `IoT` `Embedded` `Medical Devices`
🦠 CNN for COVID Detection Convolutional neural network predicting COVID infection from medical imagery. `PyTorch` `CNN` `Medical AI`	🎙️ Emotion via Voice — 🏆 Best Paper Non-linear signal-processing approach to voice-based emotion classification. Best Paper · I3SET 2k19. `MATLAB` `DSP` `ML`

_{More on github.com/sandipanseal 👉}

🧬	Utilizing Convoluted Neural Networks to predict COVID Infections in Individuals _{Turkish Online Journal of Qualitative Inquiry · Jul 30, 2021} View publication ↗
🎙️	Classification of Emotions using Voice Signal-A Non-Linear Signal Processing Approach _{SSRN · Jan 24, 2020} View publication ↗
🌾	IoT Based Smart Farming System _{Ideal International E-Publication Pvt. Ltd · Apr 1, 2018} View publication ↗

🤖 GenAI / LLM / Agents	📊 Data · Engineering · Testing	🔒 Security · Foundations
Claude Code in Action — Anthropic · Apr 2026 Claude Code 101 — Anthropic · Apr 2026 Introduction to Claude Cowork — Anthropic · Apr 2026 AI Agent Fundamentals — Databricks · Feb 2026 Generative AI Fundamentals — Databricks · Feb 2026 AI Security Fundamentals — Databricks · Feb 2026 Fine-tuning Large Language Models — Hugging Face · Feb 2026 Fundamentals of LLMs — Hugging Face · Feb 2026 Fundamentals of MCP — Hugging Face · Feb 2026 Automation in Production — Hugging Face · Feb 2026 AI Engineer for Data Scientists Associate — DataCamp · Jan 2026	SQL Associate — DataCamp · Feb 2026 Frontend Developer (React) — HackerRank · Feb 2026 Problem Solving (Intermediate) — HackerRank · Feb 2026 Machine Learning with Python — Level 1 — IBM · Nov 2023 Functionize AI Testing Certification — Functionize Introduction to Programming in C — NPTEL · Oct 2018	Certified Ethical Hacker (CEH) — EC-Council · Dec 2020 Ethical Hacking — ISOEH — Indian School of Ethical Hacking · Dec 2020 .Net Training — Persistent Systems

📜	Indian Patent — IN 202031037602 A · An Automated System Having Bag Valve Mask · Issued Sep 25, 2020
🏆	Winner — Conquer COVID-19 Hackathon · Institute of Innovation Council · May 2020
📄	Best Paper Award — I3SET 2k19 · JIS College of Engineering, Kalyani · Dec 2019
📄	Best Paper Contributor Award · NCMR, Indian Innovation Council, JIS University · Jul 2020
🌟	High Five Team Award (×2) · Persistent Systems · Jun 2023 & Sep 2023
🎓	Swami Vivekananda Merit-Cum-Means Scholarship · Govt. of West Bengal · 2018–2021

🎓 M.Sc. Digital Engineering (Computer Science) — Otto-von-Guericke University Magdeburg
_{Apr 2024 – Oct 2026 (expected) · Magdeburg, Germany}
_{Coursework: Human-Centered AI · Explainable AI · Recommender Systems · Reinforcement Learning · Information Retrieval · Advanced Database Models · Sensor Networks · Simulation · Business Informatics · Computer-Supported Diagnosis of Cardiovascular Diseases · SDLC}

🎓 B.Tech — Electronics & Communication Engineering — Narula Institute of Technology
_{Aug 2017 – Jun 2021 · Kolkata, India}
_{Coursework: Data Structures · DBMS · Operating Systems · Computer Networks · Machine Learning · OOP · Digital Electronics · Microprocessors}

🏫 Higher Secondary — Science — Govt. Sponsored Multipurpose School (Boys) Taki House · 2005 – 2017

"Forward-deployed engineering is where models meet reality — that's where I love to live."

Navigation Menu

Sandipan Seal sandipanseal

Achievements

Achievements

Highlights

Block or report sandipanseal

🧠 About Me

🎓 Master's Thesis — Explainable RF Jamming Analysis with an LLM Evidence Interpreter

📊 DeliveryLens-AI

🛡️ AegisOps-AI

🚦 InferOps-AI

⚖️ AI-Compliance-Gatekeeper

🧪 Multi-Agent QA — CrewAI

🏛️ Mine@TENJI — Legal Search

🌐 Company Website Clustering

🔬 Embedded HW + Cone Detection

🫁 Automated Bag Valve Mask — 📜 Patent

🦠 CNN for COVID Detection

🎙️ Emotion via Voice — 🏆 Best Paper

Pinned Loading

Uh oh!