AI in the roomA WORKING CONVERSATION
Clients, care and the therapeutic relationship.
A workshop hosted by Ariveria
AI in the room · Singapore · 14 September 2026
Understanding AI and Technology in Mental Health Practice: Clients, Care and the Therapeutic Relationship
An evening of practical questions, evidence and conversation for mental health professionals. We will move between short explanations, fictional cases and discussion in two groups.
Use this guide at your own pace. Everything discussed in the room is available here to revisit. A device is helpful, but not required to participate.
FICTIONAL CASE · MAYA
“I’ve started talking to AI at night. It helps me put things into words.
I brought a summary today because I didn’t want to forget what happened during the week.”
What would you want to understand first?
Maya, a fictional adult client, brings something new to a session.
UNDERSTAND
“Digital mental health” describes many different things.
Digital workflows
Digital workflows: appointments, forms, reminders and self-recording.
Purpose-built tools
Purpose-built tools: a defined task, audience or care workflow.
General-purpose generative AI
General-purpose generative AI: responses across many types of request.
Sources: World Health Organization · 2024. Sources are also available in the field guide.
An online form or reminder may contain no AI. A purpose-built tool may use rules, generative models, or both. A general-purpose chatbot may be used for emotional reflection without having been evaluated for that purpose.
These are overlapping descriptions, not a ranking of quality. Ask what the tool is intended to do and what evidence matches that particular use. A product label cannot answer those questions on its own.
UNDERSTAND
Its accuracy and usefulness still need checking.
Message
A person supplies words, and sometimes other material, for the system to work with.
Sources: World Health Organization · 2024. Sources are also available in the field guide.
A language model generates a continuation using patterns learned during training and the context available for the current response. The surrounding app may supply instructions, stored context, retrieved material or tools.
Memory, web access and data use vary between products and arrangements. Do not assume a fluent answer was fact-checked, that every tool searches the web, or that every conversation changes the model in real time.
An invented or inaccurate detail is sometimes called a hallucination. Confidence in the wording does not tell us whether a detail is correct.
UNDERSTAND
The same tool can serve different purposes.
Possibilities to explore, not motives to assign.
Sources: American Psychological Association · 2026. Sources are also available in the field guide.
Treat these as possibilities to explore, not motives to assign. The useful question is what the tool does for this person, in this situation, and what happens after they use it.
For a regular AI user, a further question is whether their experience changes across models, settings or conversation context. A screenshot is only one moment in a wider pattern of use.
DISCUSSION 01 · 10 MINUTES
Understand the use before deciding what it means.
Fictional case
“I’ve started talking to AI at night. It helps me put things into words. I brought a summary today because I didn’t want to forget what happened during the week.”
Discuss together
What is AI doing for Maya, and what would you want to understand before responding?
Would your questions change if this were a general chatbot, a structured journal, or a tool claiming to deliver treatment?
Use fictional or general examples. These notes are not submitted.
READ THE CLAIM. THEN REVEAL THE CONTEXT.
Feeling supported, symptom change and safety answer different questions.
Sources: Heinz et al. · 2025 · Shoshani et al. · 2026 · Cheng et al. · 2026. Sources are also available in the field guide.
Open each study panel to see its context. The purpose is not to reach a single verdict on AI. It is to practise matching a claim to the evidence that actually supports it.
A favorable result for a particular system does not validate a different app. A safety evaluation and a symptom trial also measure different things; their findings should not be combined into an invented overall score.
AGREEMENT AND REFLECTION · 2026
In three preregistered experiments with 2,405 participants, overly agreeable AI made people more certain they were right and less willing to repair a conflict. They still preferred and trusted those replies.
Limit: The studies measured intentions and ratings in experiments, not real relationships or psychiatric dependency.
THERABOT · 2025
210 adults with depression, anxiety or eating-disorder risk used Therabot, a chatbot built for treatment, for four weeks. Their symptom scores improved more than a waitlist group’s, and the gains held at eight weeks.
Limit: The trial did not compare Therabot with a human therapist, and the waitlist group had no app at all.
Study context
Participants were recruited through symptom screening, not clinics. The result belongs to this one tool and design; it does not show that general-purpose chatbots help.
E5 · Randomized Trial of a Generative AI Chatbot for Mental Health Treatment. Publisher abstract reverified; full-text access may require subscription. Numerical endpoint optional and excluded from classroom copy.
KAI · 2026
995 Israeli students reporting distress used the Kai coaching chatbot for twelve weeks. Anxiety and well-being improved more than with group therapy or a waitlist. PTSD symptoms did not differ between groups.
Limit: A self-selected student sample, self-reported outcomes, and substantial drop-out before follow-up.
Study context
The depression advantage over group therapy was not significant once multiple comparisons were adjusted for. The company behind Kai was involved in the research, and clinicians could intervene if safety concerns arose.
E6 · Efficacy of a Conversational AI Agent for Psychiatric Symptoms and Digital Therapeutic Alliance: A Randomized Clinical Trial. Open full text verified; link, do not republish figures.
AGREEMENT AND REFLECTION · 2026
In three preregistered experiments with 2,405 participants, overly agreeable AI made people more certain they were right and less willing to repair a conflict. They still preferred and trusted those replies.
Limit: The studies measured intentions and ratings in experiments, not real relationships or psychiatric dependency.
Study context
The findings separate liking a response from what that response encourages. They do not show that every reassuring answer is harmful.
E1 · Sycophantic AI decreases prosocial intentions and promotes dependence. Final publisher/PubMed abstracts verified. Full-text arm breakdowns are not used.
RELATIONSHIP
AI can enter the relationship through what the client brings.
Sources: Darcy et al. · 2021 · American Psychological Association · 2026. Sources are also available in the field guide.
Some users report a sense of connection with conversational software. That experience deserves curiosity. It does not, by itself, establish equivalence to a human therapeutic alliance or demonstrate symptom improvement.
This triangle is an original discussion diagram. The practitioner does not automatically see the client’s account. The third connection means influence through material the client chooses to discuss.
Ask what “it understands me” means to this person: availability, language, affirmation, continuity, privacy as they perceive it, or something else.
01
What is expected, discussed or withheld in the working relationship?
02
What is shared, what comes back, and what happens next?
03
What does the client choose to bring, and how will it be examined together?
A conceptual diagram. No automatic access to a client’s account.
THE EXPERIENCE OF CONNECTION
In a study of 36,070 selected Woebot users, respondents reported a strong early bond with the app. Their sense of connection is something to understand.
Limit: Self-reported bond; no direct therapist comparison; no proof of equivalent therapy.
This developer-led observational study surveyed people within five days of starting an earlier structured chatbot. It compared ratings with separate published studies, without random assignment to a chatbot or a human therapist.
SCRIPTED EXAMPLES · NO LIVE AI
Warmth, certainty and constructive reflection can come apart.
Both replies acknowledge pain.
Scripted teaching examples, not model outputs or a clinical answer key.
Sources: Cheng et al. · 2026 · Moore et al. · 2025. Sources are also available in the field guide.
These are deliberately simplified, scripted teaching examples. They are not outputs from a real model or complete clinical responses.
Ask what each reply acknowledges, what it assumes, and what it invites the person to do or consider. The point is not to award a universal correct answer; tone, context and the person’s needs matter.
Also ask whose language or context a response may miss. Evidence of stigma in evaluated systems is a reason to examine assumptions, not a claim about every current tool.
DISCUSSION 02 · 13 MINUTES
Take Maya seriously while making room to examine an interpretation.
Sources: American Psychological Association · 2026. Sources are also available in the field guide.
Fictional case
“The AI said my colleague was manipulating me. When you ask for more context, I feel like you don’t believe me. It gets what I’m saying straight away. I’ve asked it about this on a few evenings because I’m still not sure.”
Discuss together
How would you open the conversation while taking Maya seriously and making room to examine the interpretation?
What information would change your view of whether the pattern supports reflection or keeps the concern going?
Use fictional or general examples. These notes are not submitted.
8-MINUTE BREAK
We’ll be back in about eight minutes.
Optional thought: what would you want a technology partner to understand about your work?
Stretch, step away, or continue a conversation. The field guide will remain available.
PRACTICE
To what extent can AI be used here, and to what extent should it be? Start with a specific job, not the tool in the abstract.
Intake
A draft summary
What was inferred or lost?
Administration
A general email
Is it accurate and accessible?
Psychoeducation
Clarify approved information
Is the meaning preserved?
Between sessions
Organise topics to bring back
What role is the tool playing?
These are candidate uses to evaluate, not endorsements.
Sources: Sharma et al. · 2023 · Lukac et al. · 2025 · Reddy et al. · 2026 · Singapore Ministry of Health and Health Sciences Authority · 2026. Sources are also available in the field guide.
Each of these is a candidate use to examine, not a recommendation. Each has its own data flow, its own cost of getting things wrong, and its own need for review.
The research offers two bounded examples. AI feedback helped peer supporters write with more empathy. Documentation studies show that time saved and note quality are separate questions. Neither tells you a tool is suitable for psychotherapy work.
For any use you are weighing, name who checks the result, who acts on it, and whether the client understands the tool’s role.
HELPING THE HUMAN RESPONDER
In a randomized study of 300 peer supporters, people who wrote with AI feedback expressed more empathy than those who wrote alone. They could accept, edit or ignore each suggestion.
Limit: The study measured the writing, not whether recipients felt better.
Both groups received empathy training first. The task happened outside the live platform, with harm-related posts filtered out. This was peer support, not professional therapy.
MEASURING A USEFUL WORKFLOW
In one trial of 238 outpatient physicians, the Nabla scribe reduced time spent writing notes; the DAX scribe showed no significant change. A separate five-case simulation scored human notes higher than AI notes.
Limit: These studies concern outpatient medical documentation, not psychotherapy notes.
The time metric excluded editing inside the scribe app. In the simulation, the human note-takers worked without normal clinical time pressure.
PRACTICE
Make the purpose, uncertainty and responsibility visible.
Sources: Singapore Ministry of Health and Health Sciences Authority · 2026 · Singapore Ministry of Health · 2026 · Singapore Personal Data Protection Commission · 2023 · American Psychiatric Association. Sources are also available in the field guide.
This is an original workshop discussion aid, informed by professional evaluation and governance resources. It is not a validated scale, a compliance certificate or an app approval.
Use it for one proposed use. An honest result may be “we need more information.” A privacy setting, hosting location, or consent statement alone does not resolve the full data-governance question.
In the data sketch, additional providers, storage or logs are questions to verify. They are not claims about a specific vendor. Local references are provided for follow-up with the responsible organisation or professional body.
01 · Purpose
Name one narrow use and one excluded use.
02 · Evidence
Name relevant evidence and an uncertainty.
03 · Data
Sketch input, providers, storage, access and deletion questions.
04 · Human responsibility
Name a role at each decision and a fallback.
05 · Review
Name an observable outcome, a review point and a stopping condition.
Name a review point and what would make you pause.
DISCUSSION 03 · 15 MINUTES
Take one proposed use and decide: what would have to be true before you would consider it?
Sources: Singapore Ministry of Health and Health Sciences Authority · 2026 · Singapore Ministry of Health · 2026 · Singapore Personal Data Protection Commission · 2023 · American Psychiatric Association. Sources are also available in the field guide.
Fictional case · Group A
A clinic documentation proposal
A fictional clinic is considering a tool that records encounters and produces draft notes. Its demonstration is convincing. The supplier says clinicians can review before saving. The clinic has not established the full data flow, recording arrangements, evidence for its own workflow, or what happens to recordings and drafts after deletion. Some team members expect a time saving; another worries that review could become a quick skim.
Fictional case · Group B
A between-session reflection proposal
A fictional practitioner is considering a reflection tool that lets clients record thoughts and produce a summary to bring to appointments. It may help a client remember the week. It is unclear how the tool handles uncertain interpretations, what it retains, and whether clients might expect their practitioner to see or respond to entries. The practitioner has not identified evidence for the particular tool and intended clients.
Discuss together
Should this tool be used here, and under what conditions?
Group A: documentation. Group B: reflection.
If the tool changes next month, what part of your decision needs to be revisited?
Use fictional or general examples. These notes are not submitted.
RETURN TO MAYA
“I’ve started talking to AI at night. It helps me put things into words.”
Use the remaining questions to distinguish what we can answer from the material, what depends on an individual context, and what needs further verification.
This is a reflection on your own learning, not a claim that attendance certifies competence in using AI clinically.
TAKE THIS WITH YOU
Leave with something small enough to use.
An original workshop discussion aid, not a validated scale, compliance certificate or app approval.
Private next steps
Use fictional or general examples. These notes are not submitted.
Prompts to adapt using professional judgment; not a clinical assessment protocol.
Sources: American Psychological Association · 2026 · Singapore Ministry of Health and Health Sciences Authority · 2026 · American Psychiatric Association. Sources are also available in the field guide.
The field guide includes conversation openings, the five questions and direct links to research and professional resources. Keep it whether or not you choose to connect with the hosts.
Examples of conversation openings are prompts to adapt using professional judgment; they are not a clinical assessment protocol.
Research checked 13 September 2026
Read the studies and guidance behind the discussion. Findings, advice and original workshop prompts are labelled separately.
Start here
E1 Sycophantic AI decreases prosocial intentions and promotes dependence
Myra Cheng, Cinoo Lee, Pranav Khadpe, Sunny Yu, Dyllan Han, Dan Jurafsky · Science / American Association for the Advancement of Science · 26 March 2026 · Model evaluation and three preregistered experiments · Research
A response that clients prefer can still narrow reflection. Ask what the advice encourages them to believe and do, alongside how supported it makes them feel.
What this does not establish: Intentions and ratings are not actual relationship outcomes or psychiatric dependency. Do not generalize tested responses to every reassurance exchange.
Read the original ↗ · Public abstract ↗
Final publication: 3 experiments, N=2,405. Do not substitute the older preprint's 2 experiments/N=1,604.
Final abstract does not provide disclosure details; no conflict assertion is made.
Final publisher/PubMed abstracts verified. Full-text arm breakdowns are not used.
Jared Moore, Declan Grabb, William Agnew, Kevin Klyman, Stevie Chancellor, Desmond C. Ong, Nick Haber · ACM Conference on Fairness, Accountability, and Transparency · June 2025 · Therapy-guidance mapping and constructed-response experiments · Research
Check what a fluent reply assumes, omits and encourages. A polished tone does not establish that the response is appropriate to the person's situation.
What this does not establish: Constructed tests do not estimate real-world harm prevalence. The authors' philosophical account of therapeutic alliance is distinct from their measured findings.
Read the original ↗ · Open-access full text ↗
Authors limit the work to systems resembling those tested in April 2025, not arbitrary future AI.
No conflict claim made; conference-hosted paper consulted.
Full text verified; paper states CC BY-SA 4.0. Link only in this kit.
Alison Darcy, Jade Daniels, David Salinger, Paul Wicks, Athena Robinson · JMIR Formative Research / JMIR Publications · 11 May 2021 · Retrospective cross-sectional observational study · Research
A client's felt connection with a chatbot can be meaningful to them. Explore what that connection provides without treating a bond rating as proof of equivalent therapy.
What this does not establish: Self-selected respondents and developer-led analysis. No causal symptom outcome or randomized therapist equivalence test.
Read the original ↗ · Open-access full text ↗
Earlier structured chatbot; not a study of all generative AI systems.
Authors disclosed Woebot employment, stock options or fees.
Open publisher PDF verified.
Ashish Sharma, Inna W. Lin, Adam S. Miner, David C. Atkins, Tim Althoff · Nature Machine Intelligence / Springer Nature · 23 January 2023 · Nonclinical randomized study · Research
AI can help people improve supportive writing. Keep the human able to accept, edit or reject suggestions, and distinguish a better draft from a demonstrated improvement in care.
What this does not establish: No recipient symptom or long-term alliance endpoint. Harm-related posts were filtered; peer supporters were not professional therapists.
Read the original ↗ · Open-access full text ↗
Human and automated assessments concerned expressed empathy, not the original support seeker's experience.
No no-conflict assertion; disclosure section not relied on for teaching.
Publisher abstract and indexed author-hosted manuscript verified; direct author PDF retrieval may be intermittent.
E5 Randomized Trial of a Generative AI Chatbot for Mental Health Treatment
Michael V. Heinz, Daniel M. Mackin, Brianna M. Trudeau, Sukanya Bhattacharya, Yinzhou Wang, Haley A. Banta, Abi D. Jewett, Abigail J. Salzhauer, Tess Z. Griffin, Nicholas C. Jacobson · NEJM AI / Massachusetts Medical Society · 27 March 2025 · Randomized controlled trial · Research
A promising trial supports a claim about the tested intervention, participants and comparison. It does not automatically validate another chatbot or establish equivalence to psychotherapy.
What this does not establish: Screening-defined symptoms; no randomized psychotherapy comparator. No inference to other products, acute crisis care or longer-term outcomes.
Read the original ↗ · Public abstract ↗
Keep the effect-size and percentage-recovery headlines out of the presentation.
Dartmouth-funded developer research; detailed disclosure forms not evaluated here.
Publisher abstract reverified; full-text access may require subscription. Numerical endpoint optional and excluded from classroom copy.
Anat Shoshani, Bar Gurfinkel, Ariel Kor, Yael Ben-Haim, Or Kanarek, Romi Segev, Or Shafir, Romi Arbel · JAMA Network Open / American Medical Association · 14 April 2026 · Three-arm randomized clinical trial · Research
Some newer trials include active comparisons. Examine the particular format and outcomes before treating a result against group therapy as evidence about every form of psychotherapy.
What this does not establish: Self-report; restricted student sample; roughly one-third follow-up attrition.
Read the original ↗ · Open-access full text ↗
Do not claim depression superiority versus group therapy from the unadjusted confidence interval.
Shoshani: fees/options; Gurfinkel: fees/employment at KAI.AI.
Open full text verified; link, do not republish figures.
E7 Ambient AI Scribes in Clinical Practice: A Randomized Trial
Paul J. Lukac, William Turner, Sitaram Vangala, Aaron T. Chin, Joshua Khalili, Ya-Chen Tina Shih, Catherine Sarkisian, Eric M. Cheng, John N. Mafi · NEJM AI / Massachusetts Medical Society · 26 November 2025 · Pragmatic randomized trial · Research
Test the actual workflow, including the review step. A reduction in one writing metric does not establish better notes or an equivalent reduction in total administrative work.
What this does not establish: One system; brief trial; not psychotherapy-specific. Metric excludes editing inside the scribe application; registration completed after commencement.
Read the original ↗ · Open-access full text ↗
Well-being endpoints were secondary.
UCLA-funded; supplementary author disclosures linked in manuscript, not audited here.
Open author manuscript verified.
Ashok Reddy, Eric Gunnink, Chelle L. Wheat, Scott Pawlikowski, Chína M. Payne, Scott Wiltz, Terrence L. Hubert, Susan Kirsh, Evan Carey, Donna Hill, Karin M. Nelson · Annals of Internal Medicine / American College of Physicians · 17 April 2026 · Cross-sectional simulated-case evaluation · Research
Review a generated note for its clinical meaning, not just its readability. A time-saving tool still needs a separate check for omissions, attribution and unsupported statements.
What this does not establish: Simulation; humans lacked normal time constraints. Not a comparison of every real-world, clinician-edited AI note.
Read the original ↗ · Public abstract ↗
Not psychotherapy-specific; complements E7's different endpoint.
VHA funding reported; detailed author disclosures not verified from abstract.
Primary PubMed abstract verified; no open full-text URL confirmed.
Kathleen Kara Fitzpatrick, Alison Darcy, Molly Vierhile · JMIR Mental Health / JMIR Publications · 6 June 2017 · Unblinded randomized controlled trial · Research
Evidence follows the studied intervention, not the word chatbot. Before applying an older finding, check the tool's design, intended use and the people involved in the study.
What this does not establish: Small, short, unblinded trial; unequal attrition. Earlier structured chatbot, not a modern open-ended generative model.
Read the original ↗ · Open-access full text ↗
Useful historical example, not a current product recommendation.
Darcy founded Woebot Labs; company funded participant incentives.
Open publisher full text verified.
P1 Artificial Intelligence in Healthcare Guidelines (AIHGle 2.0)
Singapore Ministry of Health and Health Sciences Authority · March 2026 · Official healthcare AI guidance · Singapore guidance
Sets out responsibilities for AI developers, deploying organisations and healthcare professionals, including input/output review and contextual patient communication.
What this does not establish: Guidance accompanies applicable laws, codes and organisational policy; it is not a universal statutory checklist or product endorsement.
Discuss what responsibilities remain when AI contributes to professional work.
Official 42-page PDF opened directly on 13 September 2026. Relevant printed pages: 7, 28–32. Published March 2026; launch date 10 March 2026 supported by official HSA material.
P2 Data security requirements apply to AI tools that process patient data
Singapore Ministry of Health · 4 August 2026 · Official parliamentary answer · Singapore guidance
MOH states that data-security requirements apply to AI processing patient data on cloud services or on premises and separately describes additional public-healthcare practices.
What this does not establish: Public-healthcare non-retention commitments cannot be assumed for private practices or consumer accounts. The answer does not determine a particular organisation's compliance.
Explain why hosting location or a privacy setting alone does not resolve data governance.
Official page opened directly on 13 September 2026; answer paragraphs 1–2 checked.
Singapore Health Sciences Authority · Date not stated · Official regulatory overview · Singapore guidance
Explains that intended medical purposes generally bring digital health tools within medical-device regulation and links to relevant software-device and AI guidance.
What this does not establish: Does not establish any named product's registration, exemption, effectiveness or suitability. Singapore requirements should not be replaced with US assumptions.
Ask what the product claims to do and which intended use is being evaluated.
Current official page opened directly on 13 September 2026. No specific publication date recorded; site footer date is not treated as publication date.
P4 Advisory Guidelines for the Healthcare Sector
Singapore Personal Data Protection Commission · 20 September 2023 · Official sector guidance on PDPA interpretation · Singapore guidance
Addresses healthcare data collection, use and disclosure, including consent and exceptions, protection, retention, transfers, breach notification and accountability.
What this does not establish: Consent alone does not determine lawful or appropriate use. Applicable duties depend on the organisation, activity and arrangements; this workshop does not provide legal determinations.
Map what information goes where and identify questions for the responsible organisation or data-protection lead.
Revision date and relevant topics checked against indexed text from PDPC's official PDF and landing page on 13 September 2026. Direct reads returned sparse text; full direct PDF extraction was unavailable.
P5 Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models
World Health Organization · 2024 · International health-governance guidance · Professional resources
Describes generative AI applications in health and risks including inaccurate output, bias, automation bias and cybersecurity concerns.
What this does not establish: Potential applications are not evidence that a specific tool is effective, safe or suitable. WHO guidance is not Singapore legislation.
Introduce why fluent output still needs verification for its intended purpose.
Official publication page and WHO's 18 January 2024 summary were checked on 13 September 2026. Summary URL: https://www.who.int/news/item/18-01-2024-who-releases-ai-ethics-and-governance-guidance-for-large-multi-modal-models
American Psychiatric Association · Date not stated · Professional app evaluation framework · Professional resources
Six steps: Background; Access; Privacy and Security; Clinical Foundation; Usability; Integration toward Patient-Centered Goals.
What this does not establish: Structured inquiry does not certify safety or legal compliance. The workshop's five-question aid is original and is not a validated version of this model.
Show a professional framework for asking product, evidence, privacy, usability and care-fit questions.
Official page re-opened directly on 13 September 2026. Full step headings checked; navigation abbreviates the last as Data Integration. No publication date displayed.
P7 Use of generative AI chatbots and wellness applications for mental health
American Psychological Association · November 2025 · Professional health advisory · Professional resources
Recommends against replacing qualified care with chatbots or wellness apps and distinguishes mental-health-specific tools from general-purpose chatbot evidence.
What this does not establish: US professional advice is not Singapore law. Cite individual studies for numerical effects and do not generalise all tools' capabilities or risks.
Separate feeling helped, research evidence and suitability for a particular person and setting.
Relevant text and November 2025 creation date checked through indexed official APA text on 13 September 2026. Direct page reads returned only a shell.
P8 Discussing AI use in therapy
American Psychological Association · 16 June 2026 · Professional practice article with expert commentary · Professional resources
Encourages discussing clients' AI use, understanding its role, exploring concerns collaboratively and maintaining ongoing dialogue.
What this does not establish: Expert commentary is not a validated clinical protocol or evidence of treatment effects. US psychologist survey figures do not establish Singapore client prevalence.
Rehearse a curious opening conversation about AI use using a fictional vignette.
Official-domain indexed article text confirms Zara Abrams as author and Date created: June 16, 2026. Rechecked 13 September 2026. Direct access, including the .html version, returned a one-line shell; no full direct-page read.
CONTINUE THE CONVERSATION
Keep the field guide. If you would like to stay connected, you can find the hosts here.
Your hosts
Hosted by Ariveria · Nicholas Ng and Neil Gupta
Ariveria on LinkedIn ↗
