Hack Night at Tessl

RefundHunters

The Problem Hidden fees cost UK consumers up to £3.5 billion every year. The CMA just fined the AA £4.2 million this month for drip pricing under new 2024 law. Yet most people have no idea how to fight back. They accept the charge, feel frustrated, and move on. What We Built Refund Hunter is an AI tool that empowers UK consumers to reclaim money lost to hidden fees. You describe what happened in plain English — a Ticketmaster service fee, a gym joining charge, a Ryanair baggage fee added at checkout — and Refund Hunter: Identifies the exact UK law that protects you Tells you precisely what you're entitled to Generates a professional, legally-worded refund letter in seconds Maps your full escalation path if the company ignores you How It Works We use Neo4j to store a knowledge graph connecting consumer situations to relevant laws, entitlements, and escalation paths. When a user describes their situation, our AI matches it to the right node in the graph and generates a personalised response using Kimchi's free inference. Tessl skills ensure our AI writes accurate, up-to-date code throughout. Why Now The Digital Markets Competition and Consumers Act 2024 came into force this year, explicitly banning drip pricing. Companies are still breaking it daily. Consumers finally have the law on their side — they just need a tool to use it. Tech Stack Neo4j — knowledge graph of UK consumer laws, situations, and escalation paths Kimchi — free AI inference for letter generation Tessl — agent skills for accurate code Node.js + Express — backend API Vanilla JS — lightweight frontend Real Demo We found live evidence of hidden fees on Ticketmaster tonight — £15 ticket, £20.10 at checkout, £5.10 in fees never shown upfront. That's illegal. That's exactly what Refund Hunter fights.

HackerSquadNeo4jKimchi (by Cast AI)Tessl

MoveGP

# NHSync — Healthcare That Moves With You ## The Problem When people move homes in the UK, they often forget to update or change their NHS GP registration. The process can feel confusing, time-consuming, and easy to delay until an emergency happens. This especially affects: * students * immigrants * young professionals * families moving cities * elderly people As a result, many people stay registered with a GP far away from where they actually live, making healthcare access harder when they need it most. --- ## Our Solution NHSync is an AI-powered platform that helps people quickly find and switch to the best nearby NHS GP after moving homes. Instead of manually searching and filling complicated forms, users simply: 1. enter their new address 2. verify their move 3. get smart GP recommendations 4. transfer registration easily The platform recommends GPs based on: * distance * availability * accessibility * waiting times * transport access * language support --- ## What Makes It Different NHSync uses Neo4j graph technology to understand relationships between: * people * locations * GP clinics * hospitals * pharmacies This allows the platform to provide smarter and more personalized healthcare recommendations instead of just showing the nearest clinic. An integrated AI assistant also helps users understand the process and reduces confusion during registration. --- ## Real-World Impact NHSync helps make healthcare more accessible, especially for people going through stressful life changes like relocation. It can help: * reduce delays in GP registration * improve access to nearby healthcare * support vulnerable communities * modernize healthcare onboarding experience --- ## Vision We believe healthcare should adapt to people’s lives automatically. Just like banks and delivery apps update when people move, healthcare systems should also become smarter, faster, and easier to use. ## Tagline ### “Healthcare that moves with you.” # NHSync — Healthcare That Moves With You ## The Problem When people move homes in the UK, they often forget to update or change their NHS GP registration. The process can feel confusing, time-consuming, and easy to delay until an emergency happens. This especially affects: * students * immigrants * young professionals * families moving cities * elderly people As a result, many people stay registered with a GP far away from where they actually live, making healthcare access harder when they need it most. --- ## Our Solution NHSync is an AI-powered platform that helps people quickly find and switch to the best nearby NHS GP after moving homes. Instead of manually searching and filling complicated forms, users simply: 1. enter their new address 2. verify their move 3. get smart GP recommendations 4. transfer registration easily The platform recommends GPs based on: * distance * availability * accessibility * waiting times * transport access * language support --- ## What Makes It Different NHSync uses Neo4j graph technology to understand relationships between: * people * locations * GP clinics * hospitals * pharmacies This allows the platform to provide smarter and more personalized healthcare recommendations instead of just showing the nearest clinic. An integrated AI assistant also helps users understand the process and reduces confusion during registration. --- ## Real-World Impact NHSync helps make healthcare more accessible, especially for people going through stressful life changes like relocation. It can help: * reduce delays in GP registration * improve access to nearby healthcare * support vulnerable communities * modernize healthcare onboarding experience --- ## Vision We believe healthcare should adapt to people’s lives automatically. Just like banks and delivery apps update when people move, healthcare systems should also become smarter, faster, and easier to use. ## Tagline ### “Healthcare that moves with you.” ## Real World Application NHSync makes healthcare more convenient, particularly for individuals undergoing significant life transitions such as moving. It will be able to: * reduce GP sign-up time delays * increase access to local healthcare facilities * help vulnerable communities * make healthcare sign-up process more efficient --- ## Vision We envisage that the healthcare system becomes self-adapting to the changing life of an individual. The way financial institutions and courier services adjust upon relocation, healthcare needs to become equally responsive and effective. ## Tagline ### "Healthcare that moves with you."

HackerSquadNeo4jKimchi (by Cast AI)TesslNext.js 15, TypeScript, TailwindCSS, Framer Motion, shadcn-style UI components, Zustand, React Hook Form, Zod, OpenAI API abstraction, Mapbox integration, Recharts, Lucide Icons, Docker

Daemon

Atlas: Founder Memory OS Atlas is a hackathon MVP that turns messy startup notes into a Neo4j knowledge graph, then uses Tessl-style skills to change how the AI extracts, queries, and explains company memory. What It Demonstrates Neo4j as the connected memory layer for founders. Tessl skills as versioned Markdown context that teaches AI behavior. Non-chat surfaces: graph view, blocker boards, investor concerns, roadmap pressure, and evidence trails. A small chat surface that answers from graph context instead of free-floating conversation.

Neo4jTessl

whahat

WhatsApp Memory Graph is a WhatsApp self-chat assistant that turns explicitly approved group conversations into a queryable memory graph. The assistant uses Tessl skills and guardrails to keep the privacy boundary clear: it only reads messages from the user's own WhatsApp self-chat or from groups approved by token/JID, and it scans group metadata before approval. Once a group is approved, it captures live messages with source IDs and redacted senders, extracts structured facts such as tasks, decisions, links, questions, people, and notes, and stores them in Neo4j. Users control everything from WhatsApp with commands like !groups scan, !groups approve, !today, !tasks, !extract decisions from <token>, and !alvis respond. The project is built as a Go CLI using whatsmeow for WhatsApp connectivity, Neo4j for the graph store, SQLite for WhatsApp session storage, Docker Compose for local infrastructure, and an OpenAI-compatible chat completion endpoint with a rule-based fallback.

Neo4jTessl

D

sorting hat

Tell us what you read -- and we'll map the startup you were born to build (ie which startup you should join based on your interests and personality traits)!

HackerSquadNeo4jTessl

I built a knowledge graph RAG agent. Instead of storing documents as flat chunks, the data lives as a connected graph in Neo4j. When you ask a question, the agent traverses relationships — so it doesn't just find relevant nodes, it finds what's connected to them. Traditional RAG misses those connections. Graph RAG doesn't. I used Kimchi for free LLM inference and a Tessl skill to make sure the Cypher queries worked first time.

Neo4jKimchi (by Cast AI)Tessl

A graph-powered movie relationship explorer built with Neo4j Aura. Visualises connections between actors and films as an interactive graph, making it easy to discover relationships and patterns in movie data.

Neo4j

TCR

pitch deck: https://recon-silk.vercel.app/pitch Recon turns any Python repo on GitHub into an interactive, query-able graph of its own structure. Paste a URL, hit Analyze, and within seconds you get: A graph in Neo4j — every file, function,CALLS edge and IMPORTS edge from the repo, stored in your Aura instance. Subsystem clusters — Kimchi groups the functions into 3-8 labelled subsystems (auth, sessions, adapters, ...). Each color in the graph is one cluster. ASCII dataflow — a copy-paste-friendly view of the same clusters and the weighted cross-subsystem calls between them. Natural-language chat — ask "what are the most-called functions?" and the answer comes from a real Cypher query, not a guess. Click Show query on any answer to see the underlying Cypher. currently works on localhost only.

HackerSquadNeo4jKimchi (by Cast AI)TesslVercel, Claude, Codex

L

N

Brunel Residents

Morrisons hands out paper coupons at the till every week — but most of them expire forgotten in a pocket or a drawer. Coupon Vault solves this with a dead-simple app: you take a photo of your paper coupon, and AI reads it for you instantly. No typing. No manual entry. Just point, scan, and save. The app uses OpenAI's vision AI to extract the product name, discount amount, store and expiry date directly from the coupon photo. All your coupons are saved in one clean list. Any coupon expiring within 3 days is automatically flagged in red so you never miss a deal. The problem: Millions of paper coupons go unused every year because people forget about them before they expire. The solution: A phone-friendly app that turns a paper coupon into a digital reminder in under 10 seconds. Built with: OpenAI Codex — built the entire app and powers the coupon vision scanning Neo4j — graph database storing coupons and connecting them by product category, expiry and store Tessl — custom skill teaching the AI exactly how to read Morrisons coupon formatting

HackerSquadNeo4jTesslClaude, OpenAI Codex

YOu can now check what your skills and interests can lead you to and check if you are on track with your goals whila also talking to a mentor and get to achieve your dreams. (sorry video was improvisation)

HackerSquadLovable, ChatGPT, Claude

Octuple

Every time you Google a restaurant, you're voting for what you like. But Google throws those votes away — your fifteenth search for "where should I eat" gets the same generic results as a tourist. Semantica reads your search history and builds a personal taste graph in Neo4j. The places you keep coming back to become anchors. When you search, we don't rank by what's popular — we project outward from your anchors through the graph: same dish, same cuisine, same neighbourhood, same vibe. You loved Beigel Bake on Brick Lane? You'll see Brick Lane Bagel Co., Poppies Fish & Chips, Ottolenghi Spitalfields — each with the literal graph path that earned it the spot. Google ranks by what everyone clicks. Semantica ranks by what you keep coming back to. 10-second version (the one-liner): Semantica turns your Google search history into a personal taste graph in Neo4j — so when you ask "where should I eat," the answer comes from the places you've already proven you love, not from whatever the internet clicked most this week. One-line hook (for a tweet or a judge walking past): Search that knows your spots — your favourite places become a graph, and every new search walks it. A note on which to use: the 30-second version is the one to actually say out loud — it has a problem, a mechanism, a concrete example, and a punchline. The one-liner is your "judge stops at your table for 4 seconds" version. The tweet line is what goes on the slide and the README. Neo4j (Gold) — Already the core database. The taste graph, Cypher queries, ranking algorithm, and seed data all run on Neo4j. Prominently credited in the sponsor badge. Tessl (Gold) — Created .tessl/ with two skill definitions: - venue-enrichment.yaml — defines the Claude prompt, schema, and eval criteria for tagging venues with dishes/cuisine/vibe during ingest - query-boost.yaml — defines the fuzzy matching algorithm used in both Python and TypeScript, with a changelog tracking the v1→v2 regex→fuzzy migration. Both consumer files are listed so Tessl can verify they stay in sync. Kimchi / Cast AI (Gold) — Added ObservabilityMiddleware to the FastAPI backend that tracks per-endpoint latency, error rates, p95, and throughput. Exposed via GET /api/metrics. Response times are also set in X-Response-Time-Ms headers. HackerSquad (Silver) — Added a "Share Graph" button to the profile card. Copies the user's taste graph as a formatted text block (venue list + stats + HackerSquad attribution) to clipboard, or uses native share on mobile. Credited in the sponsor badge.

HackerSquadNeo4jKimchi (by Cast AI)Tessl

MovieLink

In short, your project, CineGraph, is an AI-powered movie discovery engine designed to fix the "choice paralysis" of modern streaming platforms. Instead of relying on a rigid, traditional search bar or a standard AI that hallucinates facts, CineGraph uses an Agentic Graph-RAG architecture. Here is the breakdown of what that actually means: The Core Mechanic: It uses an AI Agent (powered by OpenAI and Tessl) to understand messy human text and translate it into a highly complex database query (Cypher). The Database: It searches a Neo4j knowledge graph of TMDB movie data, which can map the complex relationships between actors, directors, and movies instantly without crashing. The MVP / Demo: To prove how fast and powerful this tech is for the hackathon, you built a gamified feature called the "Six Degrees Speedrun." Users challenge the AI to connect two completely unrelated actors, and the AI instantly maps the shortest cinematic path between them. Ultimately, it is a prototype that proves treating media like an interconnected web (a graph) rather than a spreadsheet (a relational database) is the future of content discovery!

Neo4jcodex

Vibin' Dirty

Multi-modal supply chain disruption radar. Graph-native cascade simulation across sea freight and air cargo using Neo4j and React. Type any disruption - watch what breaks.

HackerSquadNeo4jTessl

halen

Halen is a menubar app that watches the text near your cursor and runs a set of small, focused plugins against it. Every plugin runs locally — typo correction is a static dictionary; everything else goes through your own Gemma 4 instance. The text never leaves your Mac. An agent you can trust and build on

Kimchi (by Cast AI)

TBC

Problem description: Over 90% of users accept terms and conditions without reading them (https://www.businessinsider.com/deloitte-study-91-percent-agree-terms-of-service-without-reading-2017-11). Dense contracts hide clauses that could reduce your rights as a customer and may give up personal data access to other vendors. Inspiration: We wanted to reduce hitting yet another "I Agree" popup and realising there was no practical way to know what was in it and that this is a problem an AI agent should actually be solving. Solution: browser extension that reads T&Cs from websites you visit in real-time. The agent searches and evaluates the risk level of terms and conditions by analysing the text, flags risky clauses with plain-English explanations, and asks how you want to respond.

Neo4jKimchi (by Cast AI)TesslLovable, Codex, Claude

Applegate Labs

Privacy Graph — pick the apps you use, see who they share your data with as a live knowledge graph, and get AI-generated opt-out steps to take back control.

Neo4jKimchi (by Cast AI)Tessl

N

Save My Stomach

Turning food deserts into profitable oases by guaranteeing demand for mobile market traders. 💡 Inspiration Over 1.2 million Britons live in food deserts—neighborhoods where the nearest fresh produce is over a 20-minute journey by public transport. While mobile market traders could serve these areas, driving a van full of perishable goods to low-density neighborhoods without guaranteed buyers is a financial gamble they can't afford to take. We realized the problem isn't a lack of supply; it's a lack of aggregated, zero-risk demand. ⚙️ What it does SaveMyStomach completely flips the logistics model. Instead of hoping for customers, traders only drive where they are guaranteed to make a profit. We allow residents in food deserts to pre-order and pre-pay for fresh groceries. Our system continuously aggregates this hyper-local demand and surfaces highly profitable "infill stops" to mobile traders. If a trader is already driving from Market A to Market B, SaveMyStomach calculates the exact, minor detour required to pick up hundreds of pounds in guaranteed, pre-authorized revenue, maximizing their yield per kilometer. 🏗️ How we built it We built a dual-sided marketplace with a robust, highly optimized backend: - Frontend: Next.js 16 (App Router) styled with TailwindCSS and shadcn/ui. We integrated React-Leaflet for dynamic, SSR-safe mapping so residents can view upcoming trader circuits. - The Routing Engine (Neo4j): Traditional SQL JOINs are too slow and rigid for spatial routing. We built our entire backend on a Neo4j AuraDB graph model. By treating residents, LSOAs (geographic boundaries), and traders as interconnected nodes, a single parameterized Cypher query instantly calculates aggregated LSOA demand against a trader’s 1.5x direct-distance route. - Trader Intelligence (Kimchi LLM Gateway): Traders operate in vans with spotty 5G. We integrated the Kimchi Gateway to route OpenAI SDK requests, giving us failover reliability for two critical features: a vision model for instant stock recognition and a rapid-response chat assistant to help traders manage their inventory on the fly. - Infrastructure: A fully automated CI/CD pipeline using GitHub Actions, pushing Docker images to GHCR, triggering webhooks for zero-downtime Komodo redeployments on Unraid, all proxied through Caddy. 🛑 Challenges we ran into Handling the financial liability of "no-shows" was a massive hurdle. We couldn't instantly charge residents if a trader got a flat tire, nor could we force traders to refund hundreds of individual micro-transactions. We solved this by implementing a pre-authorization model—funds are only held, not captured, until the trader physically checks into the geofenced LSOA coordinates. 🏆 Accomplishments that we're proud of - Graph-Native Thinking: We successfully bypassed traditional relational databases to build a routing engine where relationships (like `[:CONFIRMED_STOP]` and `[:LIVES_IN]`) are first-class citizens. - The Infill Algorithm: Creating a system that dynamically updates the profitability of a detour in real-time as neighborhood demand grows. - Streamlined CI/CD: Merging three separate development streams (Frontend, Backend, LLM) into a seamless, automated deployment pipeline during the time constraints of a hackathon. 🚀 What's next for SaveMyStomach We want to implement a Trader Reliability Graph. By utilizing our existing Neo4j architecture, we plan to track `[:COMPLETED]` vs. `[:FLAKED]` relationships. This will allow the algorithm to prioritize highly reliable traders, ensuring that when a food desert is promised an oasis, it actually arrives. We also plan to integrate live inventory updates, handling edge cases where a trader might run out of a specific item mid-route.

Neo4jKimchi (by Cast AI)

**Daily engineering visibility for distributed teams.** A Next.js application that turns repository activity into structured reports, an evolving **Neo4j** collaboration graph, and an **LLM-assisted** interface for read-only exploration. Built for Field Engineers who need deterministic summaries across time zones without living inside GitHub diffs all day. --- ## Why this exists Remote FDEs often anchor on US hubs while staying aligned with roadmaps that move in GitHub. Pulling signal from raw commits and PRs across many repositories is slow and easy to get wrong. This project provides: - A **repeatable daily snapshot** (commits, PRs, authors, branches, repositories) with filters. - A **graph layer** that connects people, offices, repositories, and time so you can ask structured questions later. - A **mock data path** for demos and hackathons, with a clean seam to swap in real GitHub ingestion.

HackerSquadNeo4jKimchi (by Cast AI)

y

P

Tower

Tower is a real-time news intelligence platform built for traders at every level — from someone placing their first trade to seasoned professionals managing complex positions. For new traders, the platform acts as a guided entry point into the markets. The onboarding experience assesses your interests, risk comfort, and knowledge level, then eases you in: explaining how news moves markets, what to watch, and why it matters. You're not thrown into a firehose of headlines — you're walked through them with context, so you build real market intuition from day one. For experienced traders, it's a precision tool. The platform aggregates and filters global news — geopolitical events, economic data releases, supply chain disruptions, central bank decisions, and more — and surfaces only what's relevant to your specific markets and open positions. If you're trading crude oil, you're not seeing generic energy headlines. You're tracking OPEC production decisions, Middle East tensions, US inventory data, shipping route disruptions, and weather events hitting refinery capacity — all correlated to price movement in real time. Tying it all together is a personalized skill agent that adapts to who you are and how you trade. For beginners, it teaches. For veterans, it sharpens. It learns whether you're scalping momentum, riding macro swings, or building around fundamentals — then contextualizes every story accordingly. It doesn't just tell you what happened. It tells you what it means for your next move. Think of it as a research desk that meets you where you are and grows with you

HackerSquadNeo4jKimchi (by Cast AI)TesslNextJS, react, HTML, Neo4j, Tessl product: skills Acceptance Check, Visual Direction, Architecture preferences, Onboarding Rules

KingsCross Viber is an AI-assisted night planner for King's Cross. It models nearby venues, vibes, and walk links as a place graph, then turns that graph into a believable 3-stop micro-adventure based on mood, time, budget, and walkability. Neo4j is the route structure layer, making the recommendation explainable through connected venues and walking edges. Tessl acts as the behavior contract, defining how recommendations should balance vibe match, plausibility, distance, and one surprising fact. Kimchi is the lightweight narration layer that turns the graph result into a concise concierge-style explanation instead of a raw ranked list.

Neo4jKimchi (by Cast AI)Tessl

LeaseHunter

LegalGraph turns your tenancy agreement into an interactive knowledge graph. Upload a lease, and AI instantly maps every clause, obligation, and fee — flagging illegal charges like blanket cleaning fees or oversized deposits under the Tenant Fees Act 2019. Click any node for a plain-English explanation, then generate a formal dispute letter to your landlord in one click. Landlords have solicitors. Now tenants have LegalGraph.

HackerSquadNeo4jopen ai

Agent Skill Studio

A real skill-transfer platform. A Master Agent teaches. A Student Agent learns. The Evaluator proves improvement with before/after scores and a portable skill artifact. I do a before and after analysis of the agent. I analyze its response to the same bit of question after teaching it a skill and test its delta to the before skill analysis result.

HackerSquadNeo4jKimchi (by Cast AI)TesslElevenLabs

CareerGraph is a graph-first AI career mentor demo. It turns a learner profile into a structured career graph, recommends a next skill, selects a portfolio project, and generates an HTML career map. CareerGraph is an AI-powered career mentor that turns a learner’s background into a visual knowledge graph. Instead of giving generic advice, it maps their skills, evidence, projects, missing knowledge, resources, and target career into Neo4j. The system identifies what the learner already knows, what they are missing, and the next best skill to learn. It also recommends a portfolio project, learning resources, and a mentor persona, then explains the recommendation using graph paths. Built with Python, Neo4j, an LLM extraction layer, and an Obsidian-style HTML graph report, CareerGraph makes career planning more structured, visual, and explainable.

Neo4jTessl

Env_Agent

Setting up enviornment variable can be very hard for first-time coding user, this agent can control mouse and open up the environment variable window to guide the user really step by step. This can really help first-time user

A Cloudflare Worker that helps failing startups find their next move. You describe what your startup does and what's not working, and it generates a tree of 5 pivot suggestions ranked from "pragmatic refinement" to "fully unhinged but technically feasible." How the three pieces fit together: Kimchi — The AI inference layer (OpenAI-compatible API). It does two jobs: first extracts the startup's transferable assets (e.g. social-graph, payment-infrastructure) using gpt-4o-mini, then generates and evaluates the pivot tree using gpt-4o. The whole pivot exploration runs as a 3-level tree (6 root pivots → 6 children each → 6 grandchildren each) streamed back via SSE. Neo4j — The graph memory. Before generating pivots, it's queried for historical pivot patterns that match the startup's assets (findPivotPatterns). After generation, the entire explored pivot tree is persisted back to Neo4j (storePivotTree) so future startups with similar assets can benefit from past patterns. Tessl — The spec layer (the .spec.yaml file). It declares the input/output schema, constraints (e.g. desperation levels must be strictly 1–5, one of each), and model config in a structured format essentially the contract the worker implements.

HackerSquadNeo4jKimchi (by Cast AI)Tesslfp, Link to Hubble, Cloudflare Worker, Hono.js

t

Tessl

Rentor

Rentor is a privacy-first, agentic RAG prototype that helps UK tenants challenge unfair landlord deposit deductions. It analyses tenancy documents, performs per-deduction hybrid retrieval, generates evidence-backed dispute responses, and uses evaluator agents to review legal grounding, evidence use, risk, and tone before producing a draft landlord email.

Neo4jcodex , RAG pipeline

Phoenix

Product War Room transforms fragmented product, engineering, customer, and revenue signals into a live causal intelligence layer that helps leadership teams understand not just what is happening across the business, but why it is happening and what to do next.

HackerSquadNeo4jKimchi (by Cast AI)TesslLoveable

AI pronunciation coaching with phoneme-level analysis, articulatory feedback, and personalised practice.

Neo4jKimchi (by Cast AI)Tesslfastapi

Pietro

An open-source relay system that lets you chat with local AI models through a web browser — from any device. No cloud APIs, no subscriptions, completely free. Project started last week as a way of helping adults that have small to no technology background in organize and manage all their documents and photos, which are currently spread across multiple platforms (mostly google drive, photos) and devices and portable usb sticks and hard drives! I forgot to record the code in the video!

RightOn

Demo: https://hel1.your-objectstorage.com/hackersquadcontent/project-recordings/project_rec_cmp49p2640082o80kk87gnzje-2026-05-13T184640.mp4 When someone asks an LLM "what's the best X", your brand is somewhere in a hidden citation graph — or it isn't. Most founders have no idea where they sit in it. Cited crawls it. Paste your URL: we fan buyer-intent queries across every model we can reach, parse what gets mentioned and what gets cited, and write the whole thing into Neo4j as a graph you can actually query. Then we hand you the fix — landing page rewrites AI can parse, content gaps your competitors own, outreach worth sending. Kimchi makes it free. Tessl keeps the agent honest

HackerSquadNeo4jKimchi (by Cast AI)Tessl

Introverts

https://github.com/AKillerPanda/InterestMap

Neo4jKimchi (by Cast AI)Tessl

ZEX101

Agent Autopsy is an autonomous AI-powered code surgeon that takes a broken repository, diagnoses its problems, heals it intelligently, and records the entire healing journey with complete transparency. Given a buggy codebase (in this demo: a FastAPI Todo API with failing tests, crashes, and security issues), the agent runs tests, performs root cause analysis using Tessl Skills, generates minimal fixes, applies them safely via git, and logs every decision, bug, patch, and test result into a Neo4j knowledge graph. This creates a rich, visual provenance trail — allowing anyone to explore why each change was made. Built for the Tessl London Hackathon, it beautifully combines Tessl (expert library skills), Kimchi (unlimited long-running AI), and Neo4j (graph memory & visualization) into a practical, observable autonomous debugging system.

HackerSquadNeo4jKimchi (by Cast AI)TesslTessl + Kimchi + Neo4j

just me

An autonomous agent that watches your CI history, identifies tests flipping pass-fail without code changes, and opens quarantine pull requests with @pytest.mark.flaky markers via the gh CLI. Fix suggestions come from an LLM, validated through a Python AST pass so only parseable code reaches the PR. Detects 3/3 known flakies in the bundled sample with 0 false positives. Works with any CI emitting JUnit XML.

HackerSquadTesslPython, pytest, JUnit XML, gh CLI, OpenAI API, hatch

Cold Start

preflight is a CLI tool that intercepts vague prompts before they reach your coding agent. It scores your intent, asks targeted clarification questions, retrieves your team's past architectural decisions from a knowledge graph, and locks a structured spec that your agent can execute without ambiguity. Built with Tessl for skill-verified context injection, Neo4j for persistent decision memory, and Kimchi for fast dual-pass LLM inference. The result: fewer hallucinations, less wasted tokens, and agents that actually know how your team builds software. Landing Page: https://preflight-cli.vercel.app

HackerSquadNeo4jKimchi (by Cast AI)TesslCodex, Exa.ai

HackerMadness

This is a project built using Python, we used Sponsor A, Sponsor B and Sponsor C , did X Y Z with tool D X D