Context
Most AI initiatives fail because they start with a chatbot instead of a workflow. The useful version starts with the work: what information the team needs, which tools hold the truth, which actions are safe to automate, and where a human still needs to approve.
I build AI systems that sit on top of real business data: Airtable, Notion, Slack, Google Workspace, CRM records, call transcripts, documents, and internal databases. The goal is not novelty. The goal is to make a team faster at the work it already does: research, screening, reporting, follow-up, analysis, and decision preparation.
Inside 
Iterativeiterative.vcEarly-stage VC fund investing up to US$500K in startups twice a year, working closely with founders for 3 months — 70+ portfolio companies across Southeast Asia and South Asia.
Iterative, this meant building production AI tools for a venture team: Slack-native MCP workflows, investment memo generation, founder voice interviews, portfolio search, and agent systems that survived daily use.
What I Build
- Slack-Native AI Assistants: Internal assistants that answer questions, retrieve records, route tool calls, and operate inside the team's existing Slack workflow.
- MCP & Tooling Layers: Model Context Protocol systems that let AI work across Airtable, Notion, Slack, Google Workspace, CRM, databases, and custom business tools without hardcoding every integration.
- Research & Memo Systems: Workflows that turn raw notes, call transcripts, application data, documents, and external research into structured briefs, memos, and decision support.
- Voice & Conversation Agents: Voice agents for intake, screening, qualification, feedback capture, and structured follow-up, with transcripts and outputs written back into the source system.
- Build-vs-Buy Evaluation: Practical assessment of vendor platforms, direct API builds, agent frameworks, latency, cost, data control, and operational risk.
- Production Guardrails: Rate limits, cost caps, approval gates, tracing, evals, human review, tool-call validation, and narrow scopes for workflows that can fan out quickly.
Stack & Use Cases
MCPmodelcontextprotocol.ioModel Context Protocol — the open standard for connecting AI models to external tools, data sources, and systems. The backbone of production agent integrations.MCPPythonpython.orgThe lingua franca of data and AI engineering — used here for ML pipelines, voice processing, and backend services.Pythonpyannote.audiogithub.comOpen-source speaker diarization toolkit built on PyTorch — identifies who spoke when in audio recordings, the first stage of any voice cloning pipeline.pyannote.audioElevenLabselevenlabs.ioAI voice platform — text-to-speech, voice cloning, and conversational agents with 5,000+ voices across 70+ languages.ElevenLabsVapivapi.aiDeveloper platform for building, testing, and deploying real-time voice AI agents over phone and web.Vapin8nn8n.ioSource-available workflow automation — node-based, self-hostable, and the workhorse behind most of the operational automations I ship.n8nAirtableairtable.comSpreadsheet-database hybrid — the operational data layer for deal flow, portfolio tracking, and internal tooling at fund scale.AirtableNotionnotion.comConnected workspace for docs, wikis, and databases — where team knowledge lives and what internal tools read from and write to.NotionSlackslack.comTeam messaging platform — the front-end where automations deliver their output: alerts, digests, and natural-language queries.SlackSlack assistantsDeal memo generationFounder screeningVoice agentsTool orchestrationTranscript workflowsHuman approval gatesEvals & tracing
Outcome
- AI systems embedded into real operating workflows instead of sitting beside them
- Teams able to query, summarise, draft, research, and act across their own data sources
- Cleaner handoffs from conversation, transcript, or document to structured records
- Practical automation of repetitive knowledge work without losing human control
- A repeatable pattern: clean data first, narrow tools second, production guardrails before scale