Work
30+ projects for 20+ clients at Kuzushi Labs. Three flagship engagements, followed by more client case studies and product work. The way they were scoped and delivered is in How I work.
Flagship engagements
Client names are withheld under NDA.
01 · Hospitality · Guest service and operations
Guest questions answered and maintenance work coordinated
Guests can get property information, check availability and report maintenance problems through an AI assistant. It answers routine questions, guides booking requests and turns reported problems into tracked work for staff, taking over repetitive information lookup and issue recording while keeping permissions and approvals with the business.
- Work being handled
- Answering property questions, looking up availability, guiding bookings and recording maintenance issues for staff to act on.
- Delivery approach
- Connected two assistants to the client’s existing systems, with different permissions for guests and staff and a portal for reviewing actions, quality and cost.
- Guest concierge and staff assistant
- Tracked actions and approvals
- Quality and cost review
02 · Conversational AI · Financial services, healthcare, education
AI phone agents that gather customer needs and arrange the next step
AI phone agents handle routine insurance, healthcare and education calls: they collect customer requirements, capture appointment requests and arrange callbacks or handoffs. This automates information-gathering conversations that staff would otherwise conduct one call at a time.
- Work being handled
- Collecting insurance requirements, recording healthcare appointment requests and gathering information for callbacks or a specialist handoff.
- Delivery approach
- A three-person team delivered the client calling platform, then developed Zoice with configurable voices, languages and call workflows.
- 40 concurrent calls delivered
- ~1.1 s response latency
- Client went on to raise seed funding
03 · Investment research · Answers backed by source documents
Find and check company information across a million documents
Equity analysts can ask questions across more than a million company filings and earnings-call transcripts, compare financial results and open the exact passage behind an answer. The AI handles finding and assembling evidence that analysts would otherwise search for document by document.
- Work being handled
- Finding relevant company reports, comparing financial figures and collecting the passages needed to support an analyst’s conclusion.
- Delivery approach
- Delivered the research platform in six weeks, made answers link to the original passages and brought the estimated document-processing cost inside the client’s budget.
- 1M+ financial documents
- 60% lower estimated processing cost
- Answers linked to source passages
More client work
- Agentic Patient Relationship ManagementA FastAPI and Next.js patient-workflow platform: domain agents, structured action previews, confirmation, patient journeys and an intent-driven command-bar prototype.Case study
- Enterprise AI Chatbot: Indonesian TelecomTelecom customers get help with billing, troubleshooting and service activation through an AI assistant, with a route to a human when needed. It handles routine support work at roughly 300,000 conversations per month; the service operation recorded around 90% first-contact resolution and 44% lower agent-related costs.Live
- Voice AI for Automotive Customer OperationsA voice AI system for a large used-car retailer recognises customer speech and generates responses during inbound and outbound calls. The work focused on automating routine phone interactions and planning for heavy call volumes, with speech recognition adapted to the customer’s domain.Case study
- Alpha LabsCommissioned by a quantitative trading firm, Alpha Labs helps researchers and traders gather research, coordinate analysis and review computed results in one workspace. The public demonstration shows how repetitive research preparation and handovers can be organised, while strategy decisions stay with the team.Live
- ZoiceBusinesses configure AI phone agents to collect customer information and pass enquiries to the right team. In a tax-filing pilot, the agent gathered the caller’s income context and classified their needs before an expert handoff, handling the initial information-gathering conversation.Live
- Enterprise Knowledge AssistantEmployees ask one assistant in Microsoft Teams about company documents and business records across seven departments. It handles searching across separate systems and returns information the employee is permitted to see, bringing answers into the tool they already use for work.Live
- AI InterviewerAn AI interviewer conducts structured first-round conversations, asks follow-up questions and prepares assessments for the hiring team. It takes over repetitive questioning so human interviewers can focus on reviewing candidates and deeper discussions.Live
- XponentiateHealthcare recruiters use one platform to draft job descriptions, screen resumes and track candidates from application to offer. It automates repetitive preparation and screening work and replaces scattered candidate spreadsheets with a shared hiring workflow.Live
- Multi-Agent Itinerary BuilderTravellers describe the trip they want and receive a personalised itinerary combining flights, hotels and packages. The assistant compares and brings together options that customers would otherwise piece together themselves, helping them move from a long list of search results towards a booking.Live
- ONEST ProtocolJob seekers can discover and apply for vacancies across connected employment platforms, while employers and training institutes coordinate applications and placements. It connects previously separate job portals so candidates and vacancies can reach beyond a single platform.Live
Products, finance and experiments
- Kuzushi AdsA platform in development to match merchants’ product offers with relevant AI conversations. Merchants would manage listings and budgets, while AI application builders could earn advertising revenue from clearly identified product recommendations.In development
- AvashiAn experimental platform explored creating animated characters that could talk to users and remember previous conversations. It aimed to bring character generation, animation and conversation into one workflow; the project closed when mobile performance and budget constraints limited further development.Experimental
- XGenMarketing teams turn a creative brief into image and video variations they can review and refine. XGen brings generation steps into reusable workflows so each new campaign can build on previous work instead of repeating the same manual setup.Live
- SA-CCR CalculatorA risk calculator helps financial teams estimate exposure if a trading counterparty fails to pay. It combines trade data and collateral-agreement terms into reviewable calculations, making the inputs and assumptions easier to check.Live
- QuantDrillCandidates preparing for quantitative finance interviews practise maths and finance questions and receive automated scoring and feedback. The tool adjusts difficulty as they progress, supporting repeated practice without a tutor reviewing every answer.In development
- BullseyeA telecom app gives users daily tokens to predict sports, entertainment and current-event outcomes and earn rewards for correct predictions. The feature gives users a reason to return regularly, addressing the operator’s need to increase engagement.Live
- PitchPilotA live internal tool turns product information into pitch decks, demo videos, outreach copy and landing-page content. It automates drafting and media assembly that would otherwise be repeated for each product, with the resulting materials reviewed before use.Live