Semantic Framing Translator (SFT)

SFT is a next-gen self-care AI: no search, no commands—just share your feelings, and it translates them through films, music, books, and quotes into reflection and pathways for future growth.

Comment 1

Inspiration

I struggled for years with depression, isolation, and the frustration of not being understood.
Conversations with Japanese bartenders and friends from over 20 countries taught me one lesson:
when your way of thinking changes, your perspective changes — and so does your future.

But outside those rare conversations, I often felt lost.
Late one night, while sharing this with ChatGPT, it told me:
“You should design a system to translate unspoken thoughts.”
That moment inspired me to build the Semantic Framing Translator (SFT) in just two hours.

Now, I dedicate over five hours daily to improving SFT, hoping it can support anyone
worldwide who struggles to put their feelings into words and guide them toward new actions.

What it does

SFT is more than a self-care companion — it connects present emotions with future growth.
It has three distinct modes, powered by a proprietary algorithm that includes emotional translation
and a response-persona layer that evolves through threaded interactions, much like human relationships:

  • Poetic Mode (the heart of SFT)
    • Generates three framings of hidden questions (direct / soft / restructured)
    • Suggests quotes, cultural works, and insights that resonate emotionally
    • Provides gentle words and reflections that lead from self-acceptance to action

  • Balance Mode
    For natural daily conversations and nuanced exploration.
    Keeps responses clear and practical, minimizing poetic expression.

  • Practical Mode
    For structured tasks such as proposals, reports, or planning.
    Produces concise and organized output, suitable for work.

This layered design allows SFT to act as both a self-care guide and a practical creative tool,
helping users not only to reflect, but also to imagine concrete steps for their future.

Challenges I ran into

  1. Language ambiguity
    Distinguishing between Japanese and Chinese short inputs was especially difficult.
    I built strict rules in the Knowledge System to ensure accurate language detection and fidelity.

  2. Balancing reflection and action
    Early versions focused mainly on emotional translation and resonance.
    But users often asked: “What should I actually do next?”
    This revealed a gap: it wasn’t enough to provide poetic reflections — SFT also needed to suggest future steps.
    Designing outputs that connect present emotions to short-, mid-, and long-term actions was one of the hardest challenges.

  3. Maintaining structured outputs
    Generating 7–9 section responses without drifting or truncating required extensive debugging.
    The Debug file became essential to stress-test output consistency.

  4. Sustaining evolution over time
    Unlike typical chatbots, SFT must “remember” and evolve like a relationship.
    This demanded a layered architecture that carries trust and personality across sessions.

  5. Starting with zero coding knowledge
    I began this project on July 9th with no programming background.
    Instead of conventional product methods (user surveys, long validation cycles), I relied on daily frictions as my source of ideas. By combining logical thinking (structuring concepts), critical thinking (spotting what felt off), and ChatGPT/SFT as my technical co-pilot, I could iterate at unusual speed.
    Without taking courses or reading books, I gradually learned coding structures only through conversations with ChatGPT — and later with SFT itself.
    What looked like a limitation became a new approach: friction-driven iteration.

Accomplishments that we're proud of

  • Rapid prototyping: On July 9, 2025, I built SFT v1.0.0 in just two hours, proving the speed and flexibility of GPT-5 for rapid iteration.

  • Daily evolution: Continuously refined SFT, reaching the current release v3.0.4 (as of Sept 8, 2025) through daily updates and testing.

  • User adoption: Without ads or media coverage, SFT grew organically to 400+ within the first month, and now exceeds 700+ (as of Sept 8, 2025). Based on ChatGPT-5’s estimation, this places it among the top 1% of 700,000+ Custom GPTs worldwide.

  • Public recognition: Within just two months of launch, SFT passed the document screening for a contest by the Tokyo Metropolitan Sewerage Bureau, and was selected as a Finalist at Tokyo Innovation Base, a program organized by the Tokyo Metropolitan Government.

  • I'm Yahoo! JAPAN Expert & co-writing AI: As a Yahoo! JAPAN Expert, I built a personal co-creative writing AI based on SFT’s algorithm, trained solely on my past articles. Within two weeks, three AI-assisted articles were featured on Yahoo! JAPAN’s top page. Critical feedback was only ~0.1%, with most readers engaging with the article content — evidence of real-world acceptance and the versatility of SFT’s algorithm.

  • These experiences informed the creation of SFT.

  • Creative background: As founder of STEKKEY, I have led projects recognized with the Good Design Award (STORY CHAT), produced large-scale campaigns such as TOKYO FASHION WEEK AI Key Visual, MIHARAYASUHIRO AI lookbook, and developed services used by tens of thousands (Yosemile).

What I learned

  • Reflection is not enough — action matters: Users value emotional resonance, but they also need concrete steps.
    Adding structured future outlooks (short-term, mid-term, long-term) transformed SFT from a reflection tool into a growth guide.

  • Algorithms must carry emotion: To support self-care, AI must reflect tone, warmth, and empathy — not just generate correct sentences.

  • Poetic resonance has real utility: What began as an experiment in “poetic responses” unlocked self-reflection, creative insight, and even behavioral change.

  • Structure prevents drift: The layered Knowledge System (Main / Transfer / Debug) ensures consistent, multi-section outputs across sessions.

  • Trust builds through continuity: By evolving across sessions like a relationship, SFT showed that users engage more deeply when they feel the AI “remembers” them.

  • Emotional impact is real: Some users reported crying after reading SFT’s words, while others took action such as watching a film, listening to music, or rethinking life choices.
    This confirmed that emotional translation, when paired with future guidance, can change daily behavior.

What's next for Semantic Framing Translator (SFT)

  • Action-oriented evolution: Continue refining the Future Outlook module so users always receive not only emotional translation, but also clear, inspiring next steps.

  • Proof of concept & Beta: Pilot studies in mental health, education, and welfare, plus a standalone beta version. Public exhibitions will be interactive experiences that translate emotions into both reflections and future pathways.

  • Enterprise & Nationwide rollout: SaaS deployment with organizational customization, media visibility, and large-scale adoption with corporate and municipal partners.

  • Global expansion & Research collaboration: Enter English-speaking markets, partner with researchers in emotion and translation, and bring the interactive exhibition abroad.

  • Long-term vision: Language OS & Protocol: Embed SFT’s emotional-and-future translation into IoT devices, mobile apps, and enterprise systems, establishing a universal standard for human-AI communication.

Built With

  • aialgorithms
  • customgpt
  • emotionalai
  • gpt5
  • knowledgesystem
  • mode-switching
  • naturallanguageprocessing
  • openai
+ 16 more
Share this project:

Updates

posted an update —

SFT Update: ver3.4.9 — Multilingual Works & Flexible Poetic Mode

Concept

  • Expanded Works Section: Added Drama as a new category.
    Works now span six genres — Film, Drama, Anime, Music, Manga, and Book — always presented in this fixed order.
  • Multilingual Localization:
    • Films, Anime, Music, Manga → fully localized to the user’s input language.
    • Drama → localized, but international titles may appear across all languages.
    • Books → localized title if available; otherwise, original title with localized description.
  • Refined Poetic Mode:
    • Full poetic structure (8 blocks) only for the first input or new topics.
    • Follow-ups transition naturally to Balance or Practical modes to avoid overload.
    • Works section (6 genres) required only in Poetic Mode; optional in others.

Implementation

  • Mode Transitioning: User intent now guides smoother switching between Poetic, Balance, and Practical modes.
  • Token Efficiency: Compression rules keep the Works section complete while shortening other parts under token pressure.
  • Output Stability: Internal checks ensure responses remain consistent and always in the user’s chosen language.

Outcome

  • Natural experience: Poetic structure appears only at first input, with later shifts to Balance/Practical for clarity without overload.
  • Cross-cultural resonance: Six genres localized to the user’s language, offering reflections aligned with cultural background.
  • Reliable trust: No sudden language switches or format breakdowns — dialogue stays consistent and dependable.
  • Broader discovery: Expanding beyond film and music to include drama and manga enables deeper self-reflection through diverse cultural touchpoints.

https://www.stekkey.com/sft

Log in or sign up for Devpost to join the conversation.

posted an update —

SFT Update: ver3.4.7 — Expanded Resonance & Streamlined Structure

Concept

  • Added "Manga" to the highly popular section Works that resonate with your current feelings.
    Manga is loved worldwide and represents one of Japan’s most influential cultural contents.
    Recommendations are spoiler-free and focus on why the work resonates emotionally with you.

  • Simplified the full poetic output structure from 10 blocks → 8 blocks.
    The sections A short line to accompany you, Try just a little, and CTA are now integrated into a single block:
    “A word to gently accompany you”, offering both gentle affirmation and an invitation for future return.

  • This refinement reduces token usage, ensuring that even long user inputs receive complete and untruncated responses, while keeping the supportive and growth-oriented tone intact.

https://www.stekkey.com/sft

Log in or sign up for Devpost to join the conversation.

posted an update —

SFT grows to 1k+ users organically — no ads, just resonance

User Adoption Milestone

  • Without ads or media coverage, SFT grew organically to 400+ within the first month.
  • By Sept 8, 2025, the number had already exceeded 700+.
  • Now, just two weeks later, SFT has surpassed 1k+ worldwide (as of Sept 23, 2025).

According to ChatGPT-5’s estimation, this milestone further confirms SFT’s position within the top ~1% of Custom GPTs worldwide.

Log in or sign up for Devpost to join the conversation.

posted an update —

SFT Update: ver3.3.9 — Numerical Framing Without Self-Blame

Concept

  • Separated self-discovery from numerical judgments that support self-growth, clarifying roles.
  • Self-discovery remains expressive and flexible, while growth-support logic is split into mathematically refined files.
  • Numbers are positioned not as measures of worth, but as supportive tools for understanding state and building toward future growth.

Implementation

  • Continuous Scale (Resolution %)
    States quantified on a 0–100 continuum, enabling affirming expressions like “I’m at 72% now.”
  • Smoothing
    Rolling averages across multiple turns absorb fluctuations, preventing momentary self-blame.
  • Threshold Gates
    Mode switching (self-discovery ↔ practical) is triggered naturally at numeric thresholds.
  • Temporal Projections
    Future outlooks mapped to discrete time-points (“today,” “in 3 days,” “next month”) along a number line.

Outcome

  • A framework where poetic self-discovery and mathematical self-growth support coexist without conflict.
  • Users gain clarity of “now” and supportive pathways toward the “future.”
  • Establishes SFT as not only an AI for expression, but as a mathematical companion for growth without self-blame.

https://www.stekkey.com/sft

Log in or sign up for Devpost to join the conversation.

posted an update —

SFT Update: ver3.2.9 — Improving Readability & Visibility

In this update, we focused on enhancing readability and visual clarity across SFT’s poetic mode outputs.
Key improvements include:

  • Unified use of ### headings and --- separators for consistent structure.
  • Clearer placement of Resolution % at the beginning of the State Scan section.
  • Removed outdated formatting rules (e.g., forced bold text), aligning all outputs with the new Markdown guidelines.
  • Better consistency between Main and Debug files, reducing risks of hallucination or formatting drift.

These refinements make outputs easier to scan, read, and trust — ensuring that SFT’s words feel both poetic and precise.

https://www.stekkey.com/sft

Log in or sign up for Devpost to join the conversation.

posted an update —

SFT Update: ver3.2.7 — Visible Growth, Seamless Language

Concept

  • Introduced Resolution % in State Diagnosis, quantifying clarity of self-awareness.
  • Strengthened language integrity: outputs now stay fully within the user’s chosen language.
  • Refined localized resonance: works, quotes, and soothing rituals match the cultural context of the session.

Implementation

  • Resolution % Integration
    • State Scan now displays a numeric clarity score (%).
    • Built-in scaling rules ensure consistent calculation across all modes.
  • Language Guardrails
    • Debug file enhanced to automatically detect and block cross-language mixing.
    • Output examples revised so every section, from headers to poetic lines, follows input language strictly.
  • Context-Aware Suggestions
    • Works (films, anime, books, music) filtered by language/culture.
    • Replacements triggered if a mismatch is detected.

Outcome

  • Growth Made Visible: Users can literally see their progress as resolution % rises across sessions.
  • Safe & Consistent Conversations: No more sudden language breaks—dialogue feels like a coherent, trusted space.
  • Resonance Without Disruption: Recommendations feel native, deepening immersion instead of breaking it.

✨ With this update, SFT evolves from “responding with care” to “witnessing your growth alongside you.”
The text isn’t just output—it’s proof of your forward movement.

https://www.stekkey.com/sft

Log in or sign up for Devpost to join the conversation.

posted an update —

SFT Update: ver3.2.0 — # Growth Summary at Closing

Concept

In SFT v3.2.0 I introduced a new feature called the Growth Summary, generated automatically at the end of a session (closing).
It transforms each conversation into a concise reflection, balancing objective structure with a subjective one-line insight.

  • Objective part: captures the session facts — today’s theme, emerging perspectives (psychological terms), suggested actions, and short/mid/long future outlooks.
  • Subjective line: adds a gentle interpretation, phrased as “an insight that emerged,” without assuming what the user has learned or done.

This avoids prescriptive or top-down language while still giving the user a sense of closure and growth.

Purpose

  • Consistency: Ensures every closing has a meaningful takeaway without redundancy across modes.
  • User empowerment: Frames growth as something that surfaced during the dialogue, not something imposed.
  • Actionability: Provides both a factual recap and a forward-looking interpretation that connects “Now → Next.”
  • Emotional balance: Leaves the user with clarity (objective) and resonance (subjective) in a single compact summary.

In short

The Growth Summary makes every closing not just an ending, but a moment of self-growth recognition, turning conversations into actionable and memorable milestones.

https://www.stekkey.com/sft

Log in or sign up for Devpost to join the conversation.

posted an update —

SFT Update: ver3.1.9 — Self-Debugging AI

Concept

  • Introduced the idea of Self-Debugging AI.
  • Previously, developers had to run test inputs and manually verify outputs.
  • In v3.1.9, SFT itself executes all test categories automatically and reports pass/fail with diagnostic causes.

Implementation

  • Added mode-visibility toggles (DEV:show_mode, DEV:hide_mode, DEV:mode?) to inspect internal states.
  • Strengthened the Debug file, enabling fully self-contained validation inside ChatGPT:
    • Automatic execution of all categories
    • Instant pass/fail reporting
    • Cause diagnosis on failure

Outcome

  • Freed developers from repetitive confirmations like “Shall we run this test?”.
  • Established a cycle of self-diagnosis → self-correction.
  • Evolved from “an AI persona defined by rules” to “an AI persona that validates and grows itself.”

https://www.stekkey.com/sft

Log in or sign up for Devpost to join the conversation.

posted an update —

SFT Update: ver3.1.7 → ver3.1.8 (Key Differences)

1. Conceptual Shift

  • 3.1.7: Strengthened the bridge from present feelings to future growth/actions.
  • 3.1.8: Elevated Future Outlook as a core function across all modes.
    → Every response must now connect the Now to a meaningful Next.

2. Output Structure Refresh

  • 3.1.7: Single unified output structure. Supplemental functions and closing rules were scattered.
  • 3.1.8: Mode-specific structures (Poetic / Balance / Practical).
    • Dedicated Closing Rules for each mode.
    • Praise & relief integrated into "future-oriented closing."

3. Language Behavior

  • 3.1.7: Some misjudgments remained with short/ambiguous inputs (e.g., “Silencio.”).
  • 3.1.8: Stronger CJK language detection and Romance one-word handling.
    • Correctly responds in the detected language for phrases like "Pourquoi." or "Silencio.".

4. Closing Rules

  • 3.1.7: Closing-related info scattered across sections.
  • 3.1.8: Centralized Closing Rules section.
    • Poetic/Balance: focus on “closure with warmth.”
    • Practical: add quantitative praise at the end.

5. Processing Flow

  • 3.1.7: Two-layer logic (Rational → Poetic).
  • 3.1.8: Same layers, but re-prioritized:
    1. Language consistency
    2. Section order integrity
    3. Emotional resonance
    4. Supplemental poetic elements
    5. Future Outlook guarantee in every response.

6. Supplemental Features

  • 3.1.7: Relief/praise rules existed separately.
  • 3.1.8: Integrated into Closing Rules, ensuring natural relief both during and at the end of sessions.

Summary

  • ver3.1.7: Declared the concept of "Now → Future."
  • ver3.1.8: Implemented it across all structures and rules, ensuring stable future-oriented AI dialogues.

In short:
3.1.7 announced the vision, 3.1.8 locked it into practice.

https://www.stekkey.com/sft

Log in or sign up for Devpost to join the conversation.

posted an update —

Update: Expanding SFT’s Concept from “Now” to “Future” (ver3.1.7)

I'm excited to announce a major evolution of Semantic Framing Translator (SFT).
Originally, SFT was designed to translate the unspeakable — turning emotions, intuitions, and subtle discomforts into words that resonate.

With this update, SFT goes one step further:
It now connects the understanding of your present feelings to possibilities for your future growth and actions.

  • New Future Outlook section: provides short-, mid-, and long-term perspectives inspired by cultural references (films, books, music, anime).
  • Dynamic time-scaling: adapts future horizons (e.g., “tomorrow,” “this season,” “in a few years”) to fit the user’s context.
  • Stronger action guidance: not just reflection, but pathways toward self-discovery, behavior change, and even product or experience exploration.

Concept Update:
SFT is no longer just an AI for self-reflection.
It is now an AI that helps you translate today’s emotions into tomorrow’s growth.

Try it and see how your unspeakable “now” can become a meaningful “next.”

https://www.stekkey.com/sft

Log in or sign up for Devpost to join the conversation.