Krisp Hackathon 2025

Build. Collaborate. Push the Limits of GenAI.

📍 Location

Krisp Office, Yerevan

📅 Date

September 12-14, 2025

👥 Participants

Armenia's Top Engineers

View Challenge Meet the Jury

About the Hackathon

Krisp is hosting a 3-day engineering hackathon for 30+ participants. This is a unique opportunity to team up, take on cutting-edge technical challenges, and build working solutions that push the boundaries of what GenAI can do.

This is a hands-on engineering event focused on building real, functional systems, not pitching ideas.

The Challenge

In teams of 3-4 participants, you'll have 48 hours to build an AI-powered chat application that can interact with a proprietary knowledge base.

We're excited to see solutions that explore:

  • Techniques like RAG, GraphRAG, or multi-hop retrieval
  • Context and tool orchestration (e.g., MCP-style logic)
  • Agentic workflows, memory, and planning
  • Smart integrations with functions or tools
  • Systems designed with real-world constraints: latency, cost, and scale
Download Knowledge Base

Get access to the proprietary dataset for your hackathon project

Prizes & Recognition

Compete for substantial prizes and recognition from Krisp's engineering leadership.

🎯

Recognition

Showcase your work to Krisp's engineering leaders and get valuable feedback on your technical approach.

🏆

Grand Prize

$5,000

Cash prize for the winning team that demonstrates exceptional technical innovation and working solution quality.

🌱

Follow-up Opportunities

Opportunities for continued collaboration and development with Krisp's engineering team.

Participant Selection Process

Participation in Krisp Hackathon 2025 is by selection only, with limited spots available for Armenia's top engineers.

1

Application Submission

Every prospective participant completes an application form with personal details, technical background, engineering experience, and availability. Applicants may apply individually or with pre-formed teams.

2

Selection Criteria

Applications are reviewed based on technical background, motivation and commitment, and community values including collaboration, inclusivity, and respect.

3

Final Selection

Selected participants receive confirmation via email. A waitlist is maintained, and final participant lists are shared prior to the event.

Judging Panel

The judging panel consists of Krisp engineering leaders and invited experts with deep expertise in building real-world systems.

Karen Movsisyan

Karen Movsisyan

Vice President of Engineering and Research at Krisp

VP of Engineering and Research at Krisp, building AI-powered voice products that ship fast and scale. Leads teams that turn bold ideas into real-world impact. VP of Engineering and Research at Krisp, building AI-powered voice products that ship fast and scale. Leads teams that turn bold ideas into real-world impact through rapid iteration, strong execution, and user-first thinking. Focuses on delivering innovative AI solutions that meet real user needs while maintaining high engineering standards and scalable architecture.

LinkedIn
Melik Karapetyan

Melik Karapetyan

Senior Engineering Director at Krisp

Senior Director of Engineering with over a decade of experience in AI/ML infrastructure, cloud computing, and distributed systems. Senior Director of Engineering with over a decade of experience in AI/ML infrastructure, cloud computing, and distributed systems. He specialized in scaling AI/ML workloads, cloud-native architectures, and high-performance compute, combining technical depth with strategic leadership. At Krisp, he leads engineering teams across various engineering disciplines, driving innovation in AI adoption, compute efficiency, and large-scale system design. An IEEE Senior Member and PhD, he also advised startups and enterprises on AI/ML compute strategy, infrastructure scaling, and cloud cost optimization.

LinkedIn
Davit Davtyan

Davit Davtyan

Senior Engineering Manager at Krisp

Senior Engineering Manager at Krisp leading the AI Meeting Assistant engineering team. Experienced engineer with over a decade of experience spanning frontend systems, client applications, and AI technologies. Senior Engineering Manager at Krisp leading the AI Meeting Assistant engineering team. Experienced engineer with over a decade of experience in the field. Experience spans both frontend systems, client-side applications, voice applications, AI, GenAI, and backend development. Brings deep technical expertise in building scalable AI-powered communication solutions.

LinkedIn
Anush Martirosyan

Anush Martirosyan, PhD

Head of AI Engineering at Krisp

Head of AI Engineering at Krisp with deep experience in AI, ML, deep learning, and GenAI. PhD holder specializing in cutting-edge AI solutions and product development. Head of AI Engineering at Krisp with deep experience in AI, ML, deep learning, and GenAI. PhD holder specializing in cutting-edge AI solutions and product development. Leads AI engineering initiatives focused on building innovative machine learning systems and generative AI applications that power Krisp's communication technology products.

LinkedIn
Erik Davtyan

Erik Davtyan

Senior Software Engineer at A.Team

AI systems expert specializing in evaluation pipelines, RAG systems, and Model Context Protocols (MCPs) at production scale. Full-stack engineer with expertise in AI-driven architecture and DevOps. AI systems expert specializing in evaluation pipelines, RAG (Retrieval-Augmented Generation) systems, and Model Context Protocols (MCPs) at production scale. Has designed and deployed advanced retrieval architectures, multi-hop reasoning workflows, and context orchestration frameworks for real-world GenAI applications. Full-stack & DevOps engineer with expertise in AI-driven architecture and scalable system design.

LinkedIn

Judging Process

The judging will take place in two stages with independent scoring and structured deliberation.

Final Presentation

  • All teams present their solution live in front of the panel of judges
  • Presentations must include a working demo, architecture explanation, and design trade-offs
  • Judges ask clarifying questions to assess depth of understanding and technical complexity
  • Each judge completes a detailed scoring sheet independently
  • Organizers compile the scores into a final ranking

Final Decision

  • Ties are resolved through structured deliberation, prioritizing innovation and technical quality
  • All judging decisions are final
  • Teams can request feedback after the event

Scoring Criteria

Criteria Description Weight
System Design Architecture clarity, modularity, scalability, and robustness 25%
Innovation Creativity, novel approaches, and thoughtful trade-offs 20%
Application of GenAI Going beyond basic prompts to implement meaningful GenAI systems 20%
Functionality Working demo, correctness of responses, and end-to-end system logic 20%
Efficiency Consideration of latency, compute cost, and optimization 10%
Teamwork & Presentation Evidence of collaboration, clarity of explanation, and concise demo 5%

Scoring System

  • Each judge scores projects from 1 to 10 in each category
  • Scores are multiplied by the respective weights
  • A team's final score is the sum of all weighted scores, with a maximum of 100
  • The highest-scoring team is declared the winner
  • In the event of a tie, judges deliberate with emphasis on system design and innovation

Winners Spotlight

Celebrating the most innovative and impactful projects from the hackathon.

🥈 2nd Place

Ratatouille

Outstanding achievement!

🥇 1st Place

EXPM

Congratulations to the winning team!

🥉 3rd Place

HArm

Excellent work!

Transparency and Fairness

We ensure a professional, fair, and inspiring event through transparent processes.

  • Judges disclose conflicts of interest before reviewing projects
  • Judges score independently before group deliberation
  • All judging decisions are final
  • Teams can request feedback after the event
×

Loading image...

Project Showcase

Discover the innovative projects built during the hackathon. Each project represents creative solutions to real-world communication challenges.

IntelliChat

IntelliChat is an AI-powered chat application that revolutionizes how users interact with large-scale knowledge bases. Built for the Krisp Hackathon 2025 challenge, our solution implements advanced multi-hop reasoning and intelligent query planning to process Alphabet's extensive financial documentation.

Python FastAPI LlamaIndex Gemini React RAG

Team EXPM

Sipan Muradyan, Edvard Avagyan, Armen Gabrielyan, Tigran

View on GitHub

VelociRAPTOR

VelociRAPTOR is a next-generation knowledge base architecture built on top of the RAPTOR framework, designed for speed, precision, and advanced multi-chat context support. It extends RAPTOR by introducing key components—PDF Visual Extractor (PVE), RAPTOR, and chatRAPTOR—combined to deliver higher accuracy and lower latency. Leveraging this foundation, VelociRAPTOR enables three powerful modes of interaction with knowledge bases: precise mode for minimal false positives, deep conversation mode with long-context and multi-chat continuity, and thinking mode with advanced reasoning capabilities.

RAPTOR PDF Visual Extraction Multi-Chat Context Knowledge Base Open Source

Team Ratatouille

Rafayel Susanyan, Alexan Hayrapetyan, Tatevik Ter-Hovhannisyan

Main GitHub PVE Library Presentation

HArm Bot

This project delivers an MVP ready retrieval augmented generation (RAG) platform optimized for large scale document ingestion and query answering. Files up to 100 GB are streamed directly into MinIO via presigned URLs, triggering an event-driven pipeline with Redis and lightweight workers. As a result documents are parsed, chunked, embedded using OpenAI and stored in Postgres with pgvector for hybrid retrieval. Query flow combines vector search, BM25, reranking with JINA AI and a SelfRAG verification loop to ensure accuracy and citation integrity.

RAG MinIO Redis PostgreSQL pgvector OpenAI JINA AI SelfRAG

Team HArm

Harutyun Avetisyan, Armen Ghazaryan

View on GitHub Presentation

Factory of Stars

Our project creates chat-ready avatars of real people from a curated dataset. Once an avatar is created, its owner can interact with it naturally through Telegram. The core feature is that each avatar can dynamically query its attached knowledge base during conversations to provide grounded answers with retrieved, source-linked evidence.

RAG Telegram Bot Avatars Knowledge Base Chat Interface

Team RAGang

Epifanov Dmitry, Artem Shipitsyn, Denis Rumyantsev, Konstantin Tyukalov, Gosh Kolotyan, Emilia Atanesyan

View on GitHub Presentation

Jermocik Financial AI Chatbot

The project includes advanced Data Ingestion and Retrieval based Chatbot. Structured and Text data are handled separately by different ingesting pipelines and tools. A modern React-based frontend application for the Jermocik Financial AI Chatbot that provides an intuitive chat interface for interacting with an AI-powered financial assistant for market analysis, investment strategies, and financial planning.

Next.js TypeScript Tailwind CSS RAG Data Ingestion Financial AI

Team Jermocik

Davit Khachaturyan, Ishkhan Gasparyan, Hovhannes Baghdasaryan

Frontend Backend Presentation

AI Chatbot Platform for PHP

A modular, self-hostable AI chatbot platform designed for easy integration into existing PHP applications. This platform enables private knowledge ingestion, retrieval-augmented generation (RAG), and exposes a lightweight chat widget and a REST API. It's built to scale from single-server setups to distributed, multi-service deployments, offering a pragmatic solution for adding advanced AI capabilities.

PHP RAG REST API Chat Widget Self-hosted Modular

Team AImpact

Gor Garanyan, Vahe Jaloyan, Vahan Grigoryan

View on GitHub Presentation

Knowledge Base RAG System

Our solution uses RAG through a vector database, although we also explored a Graph RAG approach using Neo4j. The system provides advanced knowledge retrieval capabilities with both vector-based and graph-based approaches for comprehensive information access and analysis.

RAG Vector Database Graph RAG Neo4j Knowledge Base Web Service

Team Panzim

Mark Zhitomirski, Pavel Naidenov

View on GitHub Live Demo