Filipp Trigub
Multi-chain LLM Copilot for Academic Teaching and Studying
Experience
Multi-chain LLM Copilot for Academic Teaching and Studying
Infolab.ai
Build a sophisticated AI copilot to augment the students’ learning experience and provide AI-derived insights to professors.
- Build a multi-chain LLM system adapting to user needs at its own accord with a Weaviate vector DB based RAG system and evaluated it with Ragas.
- Build responsive react frontend, and backend systems handling auth, data management and auxiliary services as a RESTful API.
- Deployed and managed the app to the cloud in a production environment including the CICD via multi-stage deployment.
Audio-Summarizer
Startup
Summarize audio recordings, specifically books, into concise bullet-point style summaries.
- Converted audio recordings into text using STT, in this case Whisper and Huggingface models.
- Engineered a meta-prompt and summarization loop subdividing the recording in user designated parts (chapters) and then token-limit enforced chunks to extract a sensible and concise summary for each part as well as the entire recording.
- Build a Flutter based UI allowing users to summarize their recordings and receive the summary by mail.
- Dockerized and deployed to Vercel and Azure with Terraform and CI/CD.
AI Sommelier
Online Retailer
Build a sales bot for a wine store with consistent identification and memorization of customer preferences to provide the customer with a sommelier-like experience.
- Built an LLM agent with a recommendation tool based on a general haystack agent pattern design with a Streamlit UI.
- Implemented a specified meta-prompt to correctly and consistently identify customer preferences and remember them.
- Extracted these preferences programmatically to force usage of tool after a set number of preferences has been identified.
- Dockized and deployed to Azure with Terraform.
Telegram Search&Chat
SME
Make Telegram searchable. Allow the automated exploration of a multitude of telegram channels via a GPT-bot fed by vectorized documents found by a search mechanism.
- Retrieved chat contents from Telegram with Telethon and loaded these to a FAISS DB with OpenAI embeddings hosted in a FastAPI backend.
- Implemented question-answer pipeline with Haystack, using Dense Passage Retrieval and OpenAI generation.
- Provided a light-weight UI with user authentication to Telegram and subsequent chat window.
- Deployed the bot to Azure with Docker.
GPT-powered Service Staff Assistant
d-fine GmbH
Provide a GPT-powered chatbot using internal documents to support service staff and interact with customers directly.
- Led the design and development of a lightweight GPT-powered chatbot for service-staff support for an international client.
- Contributed significantly to the initial design, technology selection, and architecture.
- Implemented a haystack pipeline with OpenAI embeddings to optimize GPT usage and integrated CI/CD and DevOps for rapid, collaborative development.
- Spearheaded technical development and introduced agile development practices, setting guidelines and conventions for the team.
Full-stack Application Design and Development
d-fine GmbH
Design and develop an application for statistical data analysis holistically and support users on-demand.
- Designed and developed architectural extensions based on client demands and limitations.
- Successfully implemented over 500 stories in a 300k+ LoC codebase, which included a dockerized Python backend hosted in AWS with PostgreSQL and Oracle DB, as well as a JS frontend, both delivered via CI/CD.
- Ensured continuous stability through extensive unit and end-to-end testing.
- Served as Scrum Master for a team of 7 developers for over a year.
Summary
Theoretical physicist with experience in tech and strategy consulting. Proven record of converting client needs into tech architecture proposals, improving these iteratively and implementing the final solution. Previous clients, as part of full-time employment, include major banks, an international central bank and multinational companies in the pharmaceutical and cloud hosting industry.
Specialized in LLM application development, bringing complex LLM-agent systems from PoC to production.
Beyond that , I am a seasoned Python, TS & SQL developer with broad T-shaped skills including data processing, training and model implementation as well as backend development and cloud infrastructure deployment.
Proficient in analytical problem-solving, identifying trends and patterns in complex data, and developing algorithms to address business problems. Comfortable working independently or in teams, remote or on-site. I care about AI and tech because I care about bringing value to my partners and clients.
Languages
Education
Humboldt University of Berlin
M.Sc. in Theoretical Physics · Theoretical Physics · Berlin, Germany · 3.6 (GER scale 1.4)
Similar Freelancers
Discover other experts with similar qualifications and experience