Makoto Suwamoto

AI Workflows & Experiments

FIS Practice Gym: A deliberate-practice web app for psychotherapists based on the Facilitative Interpersonal Skills (FIS) framework (Anderson et al., 2009). Adaptive scenarios, scored by Claude Opus 4.6 against FIS-style rubrics, with three-layer feedback: a sharpened version of the user's response, an alternative from a different therapeutic tradition, and a "master class" showing how Yalom, Rogers, or Linehan would handle the moment. Implementation in Next.js, React, and the Anthropic API. Live at fis-gym.vercel.app. (Feb 2026)

Context Management ("Master Prompt") for Internal LLM Workflows: Maintain a ~30-page Master Prompt giving our enterprise Gemini deployment structured context on org structure, roles, approval paths, and recurring workflows. Lets colleagues generate first-draft emails, briefs, and decks that need only light edits. Periodically pruned to stay within context limits. (Jul 2025)

No-Code Payroll Reconciliation for a Major Sports Event: Used Claude with Google Apps Script to reconcile ~25,000 timesheet records for 1,200 hourly staff across two incompatible systems at a global sports broadcaster. Merged data, removed outdated entries, recalculated pay, and added formula-based checks to flag improbable hours. Helped ensure on-time, legally compliant payouts and contributed to bringing total payouts in under budget. (Jul 2024)

AI-Orchestrated Research & Financial Modeling Pipelines: A research-build-audit pipeline in Claude Code. The research layer adapts Karpathy's LLM wiki structure (cross-linked markdown with atomic facts and citations), enriched with Gemini deep-research-pro for retrieval. The builder generates Excel models programmatically from the wiki to reduce hallucination risk. An adversarial auditor runs in a separate context across 2–3 rounds per workbook. A second project runs an end-to-end research workflow with Exa, Perplexity, and Firecrawl APIs, using sub-agents for parallel annotation and a verification pass that fetches every retrieved URL before final output. (Apr 2026)

Local Semantic Search & Enrichment Pipelines (SQLite + Embeddings + LLMs): Two systems on a shared stack: local SQLite as source of truth, vector embeddings for retrieval, LLM-driven enrichment, and budget-aware API pipelines. The first is a Python CLI that ingests Substack archives, embeds each article with OpenAI's text-embedding-3-large, and supports natural-language search with author, date, and word-count filters. A discovery subcommand traverses Substack's category leaderboards and peer-recommendation graph to surface candidate newsletters. The second consolidates a ~1,500-row contact directory from multi-source inputs. Each row is enriched via search and content APIs, then passed to Claude Haiku 4.5 for theme extraction. Embeddings run on the summaries rather than raw enrichment text, feeding thematic clustering for matching use cases. The architecture forked cleanly into a sister project for a separate community. (May 2026)

Private Community Map App: Invite-only Next.js 15 / React 19 web app for a private group to crowdsource and comment on places discoverable via Google Maps. Tap any place on the map to see Google's info plus community vouches and comments. Built with Supabase (Auth, Postgres with RLS, Storage, Realtime), Google Maps JS + Places API (New), and Tailwind 4; implemented end-to-end with Claude Code on the web and deployed via Vercel. (May 2026)

Cross-Cultural Perception Pipeline (YouTube + Claude + Embeddings): End-to-end Python pipeline that turns hundreds of foreign-language travel vlogs into a structured perception study of how one country is depicted to another. YouTube Data API discovery → Claude classifier and rule-based triage → per-video structured extraction (locations, 12-dimension aspect sentiment, likes/dislikes, bilingual summary) into SQLite → voyage-3-large embeddings + silhouette-tuned KMeans + Claude-labeled clusters → single self-contained HTML report with Altair charts inlined, plus a NotebookLM-ready Markdown corpus. Locale config is parameterized so any source-language × target-country pair is a one-line swap. (May 2026)

AI-Assisted Matching for a 60-Person Alumni Event: Designed the survey and matching logic for a university alumni networking event. Collected attendees' goals, sharing vs. learning topics, conversation style, and career stage via Google Forms, then used ChatGPT o3 to tag and score responses. Assigned participants to two rounds of 4-person tables balancing shared interests with complementary expertise. (May 2025)

Main Areas of Interest: How LLMs enable adaptive free-text assessments, psychometric measurement, and deliberate practice in domains where structured practice was previously impractical.

Makoto Suwamoto is a Corporate Operations Specialist based in Tokyo with a rare combination: the strategic judgment to know when AI shouldn't touch a problem and the technical sophistication to build serious solutions when it should.

They've architected a 30-page "Master Prompt" system that functions as organizational memory, built custom Apps Script automation for complex payroll reconciliation without writing traditional code, and developed verification techniques—including adversarial "canary traps"—that catch AI failures before they compound. Their approach to context management and model limitations reflects genuine understanding of how these systems actually work.

What makes Makoto's profile worth exploring: they're quantifying 90%+ time reductions while most practitioners are still figuring out basic prompts.