A podcast back catalogue is too hard to search by topic

The problem

Listeners and producers struggle to locate a useful explanation spread across dozens of episodes.

Agent job

Create a topic taxonomy, index claims to episode/timecode and build a simple searchable archive prototype; return matching excerpts with source links and no unsupported answers.

Expected output

Episode topic index and static searchable archive prototype using the creator’s own licensed material.

Why an agent / simpler alternative

Assessment

The agent can build a topic taxonomy across episodes and implement a searchable prototype. Deterministic indexing is sufficient once terms and metadata are stable. Ordinary platform search may be enough; human producers must verify nuanced excerpts and timecodes.

Setup requirements

Items

1. Creator-authorized transcripts with stable episode IDs and timecodes

2. Clear excerpt/reuse rights and a list of excluded media

3. Local static-site environment and text-search library

4. A frozen benchmark of 15 human-written queries across five authorized pilot episodes, each mapped to one creator-verified target segment (episode ID plus start/end timecodes). Before testing, creator marks five priority queries critical and ten ordinary; freeze wording, target mapping and priority labels.

Inputs and access

1. Name

   Creator-owned episode transcripts and URLs

   Source or owner

   Prospective consenting customer or specified official publisher

   Sensitivity

   Use synthetic, public or appropriately deidentified material; confirm actual sensitivity before work

   Permission needed

   Owner authorizes scoped access and processing; verify rights and retention before receiving files

2. Name

   Existing timecodes

   Source or owner

   Prospective consenting customer or specified official publisher

   Sensitivity

   Use synthetic, public or appropriately deidentified material; confirm actual sensitivity before work

   Permission needed

   Owner authorizes scoped access and processing; verify rights and retention before receiving files

3. Name

   Approved taxonomy and content-reuse permissions

   Source or owner

   Prospective consenting customer or specified official publisher

   Sensitivity

   Use synthetic, public or appropriately deidentified material; confirm actual sensitivity before work

   Permission needed

   Owner authorizes scoped access and processing; verify rights and retention before receiving files

Proposed steps

1. 1

   Select five pilot episodes and document transcript, excerpt and branding rights.

   Expected output: Rights-cleared pilot manifest

   Approval gate: Exclude third-party audio and material without reuse permission.

2. 2

   Normalize episode IDs and timecodes; flag missing passages before indexing.

   Expected output: Clean transcript copies and gap log

   Approval gate: Do not invent timestamps or missing speech.

3. 3

   Propose a small taxonomy for creator approval. Before implementation testing, freeze 15 benchmark query strings and their human-verified target segments across five episodes; creator marks five critical and ten ordinary without seeing results.

   Expected output: Approved topic dictionary and frozen query/target/priority manifest

   Approval gate: Keep opinions and differing viewpoints in their original context.

4. 4

   Build a local search index and static interface returning source-linked excerpts.

   Expected output: Unpublished prototype

   Approval gate: No hosting, public deployment, tracking or membership integration yet.

5. 5

   Run the frozen 15 queries and log each ranked top-three result against its mapped target segment. Require all five critical targets and at least 12 of 15 targets overall in the top three; manually check episode, timecode and excerpt context for every returned top-three result.

   Expected output: Per-query benchmark report, top-three evidence, critical-pass result and disclosed ordinary-query misses

   Approval gate: Any critical-query miss, fewer than 12 overall hits, misleading excerpt, incorrect source/timecode or rights failure blocks pilot acceptance. Up to three ordinary misses are tolerated only if disclosed and accepted by the creator. Passing is limited pilot evidence, not a general launch-readiness guarantee.

6. 6

   Deliver prototype, content-update instructions and a clear maintenance estimate.

   Expected output: Handoff pack and exclusions

   Approval gate: Creator separately approves hosting, fees, access controls and publication.

Synthetic output illustration — not an executed result

Synthetic

Yes

Executed

No

Text

SYNTHETIC: Search “planning a first workshop” returns Episode 04 at 00:12:30, topic “pilot session design,” with a two-sentence licensed excerpt and source link. Confidence: human timecode check pending. No episode or archive exists for this illustration.

Human approval boundaries

1. Creator approves material, excerpts and prototype; human checks relevance and separately approves hosting, publishing and fees.

2. Human service provider reviews quality, scope and quoted fee; customer approves contract, delivery recipients and payment terms. Separately approve purchases, credentials, publication and external sends. No commitments or payments are made in this research.

Failure modes

1. Episode timecodes drift

2. Topic matches can omit contrary context

3. Third-party music or quotations may have separate rights

4. Search may add no value beyond existing platform tools

Recovery and troubleshooting

Items

1. If timecodes drift, return plain episode links until a human corrects alignment.

2. If retrieval misses a critical query or the total falls below 12/15, adjust taxonomy/indexing and rerun the unchanged benchmark, retaining prior results. Disclose up to three tolerated ordinary misses; do not relabel criticality or invent answers to hide failures.

3. If rights cannot be established, remove the material and report the coverage gap.

Unexecuted validation plan

Sample setup

Five creator-authorized episodes; 15 human-written benchmark queries mapped to 15 verified target segments with episode IDs and start/end timecodes. Creator preselects five critical queries and ten ordinary queries before results are seen; freeze query text, targets and criticality. This benchmark is proposed and has not been run.

Acceptance checks

1. All five predesignated critical queries retrieve their mapped target in the top three

2. At least 12 of all 15 queries retrieve their mapped target in the top three; up to three ordinary misses must be individually disclosed and accepted by creator

3. Human checks every returned top-three result for correct episode/timecode and non-misleading excerpt context

4. No invented answer text; rights review excludes unlicensed media

5. Do not relabel failed critical queries or alter the frozen set to reach the threshold; any revised benchmark requires fresh approval and separately reported results

Stop conditions

1. Stop if data authority or rights are unclear.

2. Stop pilot acceptance on any critical-query miss, fewer than 12 overall hits, source/timecode/context error or rights failure; do not claim launch readiness.

3. Stop if manual correction time exceeds the pilot effort budget without customer approval.

Executed

No

Costs and effort — unknown or hypothetical

Status

unknown_or_estimate

Currency

CAD

Items

1. Agent subscription/API usage: not quoted; verify current plan and commercial data terms.

2. Human intake, rights review, QA and revisions: unknown until timed pilot.

3. Storage, hosting, integration and payment-processing costs: excluded unless separately agreed.

4. Customer acquisition and support time: not measured.

Caveat

No margin or income estimate. Price hypotheses are editorial offer tests, not observed market rates; taxes, scope and effort require a real quote.

Proposed service concept — earnings not verified or guaranteed

Buyer

Independent podcaster, membership creator or small network

Deliverable

Episode topic index and static searchable archive prototype using the creator’s own licensed material.

Charging basis

Setup fee for an owned archive; creator may choose to include it in an existing membership without any promised sales uplift.

Acquisition channel

Permission-based discovery with existing professional contacts, referrals or an opt-in sample page; none executed.

Assumptions

1. Customer has this unresolved problem and values a reviewed deliverable above existing built-in alternatives.

2. A consenting pilot confirms data access, achievable quality and economically tolerable review time.

3. Provider can deliver within a written scope and supports the result.

Revenue guarantee

No

Canadian adaptation — proposed

Proposed Canadian pilot: include Canadian place/person spellings, accents and any English/French queries in the creator-approved frozen benchmark; do not assume bilingual retrieval works. Confirm rights for excerpts and hosting region if later publishing is authorized. No Canadian deployment or discoverability claim.

Limits

1. Original proposed plan; untested.

2. Problem and capability references are evidence of context only, not demonstrated end-to-end success, customer demand, willingness to pay or profitability.

3. Compatibility, workload, delivery effort and availability must be verified in a real authorized pilot.

4. The 15-query benchmark is a small agreed pilot gate, not proof of corpus-wide search quality. All five critical queries must pass; up to three ordinary misses can remain only with explicit disclosure and creator acceptance.

5. Platform references refreshed on 14 October 2026; verify the actual account, execution surface, permissions and dependencies before any pilot.

Candidate platforms — compatibility untested

1. Name

   Codex

   Required capabilities

   1. File analysis, code execution and test/report artifact creation.

   Capability source urls

   1. https://learn.chatgpt.com/docs/codex/cli

   2. https://learn.chatgpt.com/docs/artifacts-viewer

   3. https://learn.chatgpt.com/docs/sandboxing

   Workflow compatibility

   Unknown; verify in the proposed first test.

   Historical candidate name

   Codex

   Selection basis

   Editorial candidate from general capabilities; exact end-to-end fit untested.