NeedRadar is an evidence-first decision system for finding AI-solvable workflow opportunities in public discussions without confusing public interest with validated demand.
Live product: needradar-ai.xusihan615.chatgpt.site
AI strategy case deck: NeedRadar_AI_Strategy_Case_Study.pptx
The live product opens in English and includes a persistent Chinese-language switch. This repository and every attached portfolio asset are English-only.
The problem is not a shortage of AI ideas. It is the absence of a repeatable way to decide which workflow deserves evidence, a pilot, and organizational attention.
NeedRadar makes three things explicit:
- Provenance: what was said, where, when, and under which source policy.
- Distinct demand: whether a task is genuinely new or repeats an existing opportunity.
- Proof boundary: what is a public signal and what still requires observed behavior.
The strategic recommendation is simple: automate evidence preparation while keeping source rights, investment priority, and adoption decisions accountable to people.
Snapshot retrieved from the public production API on 26 August 2026.
| System output | Count | What it means |
|---|---|---|
| Stored observations | 421 | Records retained by the governed evidence pipeline |
| Public observations | 117 | Reviewable source records exposed through the public layer |
| Published opportunity briefs | 10 | Deduplicated decision objects that passed publication gates |
| Source policies | 20 | 12 active and 5 submission-only at snapshot time |
These are production-system counts, not market-validation metrics. No customer count, adoption, revenue, savings, ROI, or product-market fit is claimed.
The current snapshot is preserved in data/production-snapshot.json.
flowchart LR
A[Govern source] --> B[Screen noise, PII, access]
B --> C[Structure user, job, workaround]
C --> D[Cluster independent evidence]
D --> E[Draft opportunity and unknowns]
E --> F[Compare for semantic overlap]
F --> G{Publication gates}
G -->|Complete and distinct| H[Publish brief]
G -->|Too similar| I[Suppress duplicate]
G -->|Insufficient evidence| J[Hold for more evidence]
The public card is a decision object, not a scraped-post summary. It separates:
- the user and job to be done;
- the current workaround and desired outcome;
- the plausible AI mechanism;
- human decision gates;
- counterevidence and unknowns;
- source provenance and validation level.
| Consulting capability | NeedRadar evidence |
|---|---|
| Opportunity framing | Converts vague discussion into bounded workflow decisions |
| Operating-model design | Separates system preparation from accountable human judgment |
| AI governance | Encodes source policy, PII, access, evidence, and similarity gates |
| Product strategy | Defines validation levels from public signal to sustained outcome |
| Implementation | Runs a production evidence pipeline and a public bilingual product |
| Executive communication | States the recommendation, evidence, risks, and unproven claims separately |
My role covered product framing, evidence architecture, governance design, operating-model definition, implementation, quality controls, and portfolio communication.
This public repository is a curated portfolio edition, not a deployment mirror. It intentionally excludes production secrets, operational databases, personal data, and private administrative surfaces. It includes representative, runnable policy logic:
- src/evidence-policy.mjs — evidence completeness and source-safety gates;
- src/similarity.mjs — transparent semantic-overlap scoring;
- src/pipeline.mjs — publication, suppression, and hold decisions;
- tests/pipeline.test.mjs — executable policy examples.
Run the checks with Node.js 20 or newer:
npm test
npm run check:englishSystem / AI prepares: normalization, noise and PII flags, classification, clustering, draft structure, similarity comparison, and reason logs.
Accountable people decide: source-rights exceptions, strategic priority, acceptable failure, real-task interpretation, and scale, redesign, or stop decisions.
The boundary follows decision risk. It is not a generic “human in the loop” label.
Demonstrated: a working evidence pipeline, traceable public observations, deduplicated opportunity briefs, source-policy controls, bilingual product delivery, automated tests, and explicit claim boundaries.
Not yet demonstrated: customer adoption, measured workflow improvement, commercial value, ROI, willingness to pay, or product-market fit.
See docs/PORTFOLIO_CLAIMS.md for interview-safe wording and docs/ROADMAP.md for the proposed 90-day validation path.
| Path | Purpose |
|---|---|
| docs/AI_STRATEGY_CASE_STUDY.md | Full case narrative and recommendation |
| docs/ARCHITECTURE.md | System layers and information flow |
| docs/EVIDENCE_MODEL.md | Evidence levels and interpretation rules |
| docs/GOVERNANCE.md | Decision rights, risks, and controls |
| docs/ROADMAP.md | Proposed 90-day discovery and pilot plan |
| portfolio | Recruiter-ready presentation artifact |
| src | Representative policy implementation |
| tests | Executable policy checks |
I built NeedRadar to solve a decision-quality problem, not an idea-generation problem. The system turns governed public signals into traceable AI opportunity briefs, automatically suppresses likely duplicates, and makes the validation boundary explicit. The implementation is working in production; customer adoption and business value remain hypotheses for a bounded pilot.
The representative code and original documentation in this repository use the MIT License. Third-party observations remain governed by their original source terms and are not relicensed here. This repository contains no operational dataset or personal data.