AI · Products · Markets

AI Directions & Opportunities

Catching up after a decade · May 2026

Jevin Tan

01

From Pharmacist to AI Builder

Started as a pharmacist — trained to spot patterns in data, reactions, outcomes
Realised the analytical mindset translates: drug interactions → data pipelines, clinical checks → AI workflows
The lightbulb: building tools for myself that could help entire industries
Now building three AI products simultaneously across computer vision, agentic AI, and applied AI solutions

02

What I Actually Build

🏠

Property Intelligence

ClawOne — private AI appliance for property agents. Market data, client workflows, source-aware insights.

🧠

Computer Vision

SIMIS.AI — on-premise CCTV safety monitoring. Competition-winning. Healthcare & childcare focus.

📱

AI Solutions

Bizelev8 — multilingual AI avatar platform. Korean pilot live. Selling outcomes, not AI.

"Drinking my own champagne — built these tools for myself first, then productised them"

03

🏆 1st Place — PAN-SEA AI Developer Challenge 2025

SIMIS.AI — Competition-Winning Computer Vision

AI Singapore + AngelHack competition, Healthcare Track winner
Computer vision for medical device guidance — real-time identification & usage instructions
Demoed across healthcare, tourism and logistics
Current focus: childcare CCTV safety monitoring — the highest-stakes application

SIMIS video intelligence platform visual

SIMIS turns passive CCTV into searchable, privacy-first video intelligence

04

SIMIS.AI — Market Opportunity

📹

CCTV Cameras

Existing infrastructure

➡️

💻

Edge Device

On-premise, privacy-first

➡️

🧠

AI Models

SEA-LION multilingual

➡️

🚨

Alerts

Real-time safety alerts

1,000+

Childcare centres
in Singapore

ECDA

Compliance gap —
manual supervision only

SEA

Pan-regional
expansion potential

Open-source models — no vendor lock-in, multilingual support
On-premise edge deployment — data never leaves the building
Privacy advantage vs cloud CV: critical for childcare compliance

05

ClawOne — Your Private AI Box

📦

ClawOne Box

On-premise hardware

➡️

🤖

Multi-Model AI

Plan → Execute → Review

➡️

📱

Insights on-the-go

➡️

👤

Agent

Actionable next steps

Plug-in box — pre-flashed OS, auto-updating, zero config. Power + WiFi = live
Source-aware: every insight shows data source, timestamp, confidence level
Privacy-first: customer-owned hardware, signed updates, no standing remote access
Telegram-native — no new app to learn, works from day one

Core idea: "ChatGPT is the brain. The workflow engine is the memory and manager. ClawOne is the private box that keeps it running."

06

ClawOne — Why Now

$1.35

OpenAI loses per
$1 earned on inference

30–50%

Projected API price
increase at normalisation

3.3%

Gap: top open vs
closed model (Stanford)

AI pricing is at an inflection point — OpenAI projected to lose $14B in 2026. Token pricing will likely correct upward. Cost structures built on cheap API calls won't last.
Open-source is catching up fast — 62.8% of all AI models are open-source now. The performance gap between open and closed models is shrinking to almost nothing.
Property agents are underserved — Missing 30% of business from unanswered calls. Spending 4–8 hrs on single lease abstractions. They're not technical people. They need something that just works.

The idea: A fixed-cost box that sits on their desk. Data stays with them. No IT knowledge required. Works on Telegram from day one.

07

ClawOne — Business Model

S$69

Per month
subscription

S$199–299

Refundable
deposit

36,000+

Property agents
in Singapore

Like leasing a router — pay a small deposit, get the box. Deposit refunded when you return it. Simple.
3-month minimum term, then monthly rolling. Easy to try, sticky once it's part of your workflow.
The market: PropNex 12.6k · ERA 8.6k · Huttons 5.7k · OrangeTee 5k+ · plus independents
Growth path: Start with one agent → prove value → expand to the whole team or agency.

Why hardware? People treat a physical device differently from an app subscription. It sits on your desk, it's yours. Think: router, not another login.

08

Who Else Is in This Space

Player

Focus

AI

Memory

Privacy

MOGUL.sg

Buyer-side matching

✓

✗

Cloud

PostAI

Video content gen

✓

✗

Cloud

Privyr

Mobile CRM

✗

Cloud

PropertyGuru

Portal lock-in

✓

✗

Cloud

ClawOne

Agent workflows

✓ Multi-model

✓ Persistent

✓ On-prem

Nobody combines all four — that's the gap ClawOne fills.

🔍

Source-Aware

Every insight traced to origin

🧠

Agent Memory

Learns your playbook over time

🔒

Privacy-First

Hardware you own, data you control

📨

Telegram-Native

No new app, zero friction

09

Bizelev8 — AI Solutions Play

01

Script

Interview content + localisation

02

Generate

Avatar, voice, subtitles

03

Review

Human QA before release

04

Deploy

Learner-facing outcome

HeyGenOpenAIAnthropicElevenLabs

Multilingual AI avatar platform — training, interviews, content at scale
Korean pilot live with Kilsa — validating cross-border demand
Stack-agnostic: swap any model component without rebuilding

"Not selling AI — selling outcomes"

10

Technical Foundation

The infrastructure layer that makes all three products possible:

🧠

Multi-Model Orchestration

Plan → Execute → Review pipeline. Route tasks to the right model. Automatic fallback chains.

💾

Local Embeddings

On-device semantic search. No data leaves the box. Memory persists across sessions.

⏰

Scheduled Monitoring

Cron-based agents that watch markets, scan data, push proactive alerts 24/7.

Fallback chains: If one model degrades or prices spike, automatically route to alternatives
Source-aware confidence: Every output tagged with model, timestamp, and data provenance
24/7 agent runtime: VPS-hosted, Telegram-connected, proactive not reactive

11

What I'm Thinking About

Questions I'd genuinely value your perspective on:

01

On-prem AI vs cloud — who wins long-term?

Running models locally keeps data private and costs fixed. Cloud gives scale and flexibility. Is on-prem a real advantage or a temporary niche?

02

Are tokens the new fuel?

AI isn't infinitely scalable — compute, energy, and model quality are finite resources. Does the token-as-commodity thesis hold? Or does it break?

03

What happens when inference costs normalise?

Right now OpenAI burns $1.35 per dollar earned. When pricing corrects, do on-prem and open-source win by default? Or does the ecosystem consolidate?

04

Does hardware still matter when AI goes everywhere?

If every app has AI baked in, is there still a place for a dedicated AI box? Or does the value shift entirely to software and workflow?

12

Thanks, Weicong

Not here to pitch — just excited to catch up and hear what you think.
These are real questions from someone figuring things out as he goes.

SIMIS.AI ClawOne Bizelev8

jevin@simis.app · Jevin Tan on LinkedIn · simis.ai

13