The goal of the Omote project is to harness the potential of LLMs in creating narratives to construct rich, realistic characters embedded in fictional worlds that express themselves in realistic ways. I'm trying to achieve this by introducing an internal dialogue of "mind departments" responsible for:

• personal memories (Reliquary)
• knowledge about the world (Basilica)
• current needs (Bone Chorus)
• sexuality (Murmuring)
• goals and fears (Ventral Striatum)
• identities (Wardrobe)

The departments are augmented by a collection of personality parameters that are generated during construction of the character's personal history. They participate in an internal dialogue (overseen by the mind executive - the Warden) which in turn results in an agreement or a conflict as to what the reaction to an interaction should be.

This is my fun project that is inspired by one of my favourite games, Disco Elysium. I think its primary application should be to elevate NPCs in games.

You can read more about it here or watch a short demo here.

[ LangChain ] [ LangGraph ] [ LangGraph Platform ] [ LangSmith ] [ AWS ECS ] [ Docker ]
[ Terraform ] [ GH Actions ] [ openrouter.ai ] [ gradio ]

Navigator is an application that aims at providing high-quality pentesting service for companies that can't purchase a specialised penetration tests. By automating selected section of the OWASP Top 10 (2021) classification we reduce costs and provide our customers with complex support, guiding them from detection of vulnerabilities and weaknesses in application design (OWASP A04: Insecure Design) verifying their deployment pipeline (OWASP A08: Software and Data Integrity Failures), examining the attack surface for most common vulnerabilities (A01, A02, A03, A06, A09, A10) through explaining the nature of vulnerabilities and offering guidance in fixing and retesting them.

In summary, navigator offers affordable autonomous pentest teams at fraction of price and with expanded capabilities compared to traditional scanners.

This project is for me particularly interesting because I have an opportunity to tackle such operational issues as:

• complex agent hierarchy
• model and application observability
• multi-tenancy in multi-agent application
• provisioning and pricing
• figuring out early adoption

‍

[ Python ] [ LangChain ] [ LangGraph ] [ LangSmith ] [ gradio ] [ Prometheus/Grafana ] [ AWS ]
[ openrouter.ai ] [ HF TGI ] [ HF TRL/SFT ] [ GHA ] [ Terraform ] [ Docker ]

The goal of the Toshokan project is to aid English-speaking students in learning Japanese language. In my experience on the path to learn Japanese I realised that there are many obstacles that I haven't come across learning e.g., English or Portuguese. For example:

• Formality levels differ depending on age & hierarchy
• Exposure to language is problematic. Popular multimedia don't contain equivalent of everyday Japanese
• Kanji logographic register is very large and time-consuimg

In Toshokan I attempted to harness capabilities of contemporary LLMs to remove these barriers. I'm doing so by:

• Controlling the list of known kanji and kanji that you want to schedule to memorise
• Selecting formality level for practicing different language registers. Using LLMs capability to properly select language register for formal / semi-formal / informal situations
• Breaking down the learning process into individual lesson units with limited scope that you can exercise ad infinitum or until you are satisfied
• Including a fast chat for words and auxiliary chat for general questions in context of Japanese
• Listing and annotating all unknown kanji (kanji outside of the known kanji register) during practice conversation

You can test the app in HF spaces here.

[ Python ] [ AWS ] [ LangChain ] [ LangSmith ] [ openrouter.ai ] [ gradio ] [ GHA ] [ Docker ]