The one-time stabilization engagement is complete. The team is back to generating decks — 100+ per month, 10–15 a day, all on the tool. From the team's own review, four specific gaps stand between the current state and "90% efficiency." This proposal closes those gaps and restructures unit cost and ownership in one retainer window.
Since wrapping the stabilization deliverable, exploratory work has been underway using a new agentic harness — an R&D investment not billed here — that informs the approach to several of the items below. What the team needs now is a bounded engagement that delivers exactly the four quality improvements requested, plus the cost optimization and open-stack transfer outlined in the earlier proposal.
4
team-requested quality improvements
2 months
retainer · prepaid per month
₹50,000
fixed total · 50% upfront
The Request
The team reviewed the stabilized tool and identified four items that would take it to roughly 90% of its target efficiency. Verbatim: better and more relevant images, itinerary history and prompts after pasting the Excel, better detailed descriptions for day-wise activities, and a body font size reduction from 20 to 16 to restore alignment. Separately, the operations team asked explicitly: "Is there a way we can keep track of itineraries made and see a history?" — confirming that history and tracking are not just nice-to-haves but real operational needs.
Those four items are this proposal's core scope. They are bundled here with the cost optimization and open-stack transfer — a combination that addresses both the daily-user experience and the long-term commercial position in a single retainer window. At 100+ decks per month, the ₹55-per-deck model cost is no longer negligible — it is ~₹5,500/month and growing.
The Engagement
Five workstreams
Nothing outside the list below is implied.
A
Workstream A · Image quality
Better, more relevant images in every deck
Image relevance is what the team flags most — decks with generic photos where specific ones belong. This workstream improves both sourcing and selection.
Broaden image search across providers with relevance-weighted queries.
Per-activity image prompts that pull destination + activity-specific results.
Improved fallback chain so a provider outage does not leave slides blank.
Outcome: images that reflect the actual itinerary, not a generic destination search.
B
Workstream B · Itinerary history
History and prompt persistence after pasting Excel
Today, pasting a spreadsheet generates a deck and the input is lost. The operations team asked directly: "Is there a way we can keep track of itineraries made and see a history?" With 100+ decks per month, a record is essential.
Save pasted itinerary inputs with timestamps and generated-deck references.
Display recent history in the tool so the team can revisit and re-generate.
Preserve the original prompt/Excel data so it can be re-pasted with edits.
Outcome: never lose a pasted itinerary again; one-click re-generation.
C
Workstream C · Day-wise descriptions
Better detailed activity descriptions per day
The day-wise summaries were condensed during stabilization (1–3 lines per day). The team now wants richer detail so each day slide tells a clearer story for the client.
Expand day descriptions to include activity context, timing hints, and highlights.
Tune the generation prompts to produce more vivid, sale-ready copy per day.
Keep summaries scannable — detail without walls of text.
Outcome: day slides that sell the itinerary, not just list it.
D
Workstream D · Font alignment
Body font 20 → 16
The July layout review bumped body font to 20. In practice, this breaks slide alignment — text overflows, spacing collapses, and decks look inconsistent across PowerPoint versions.
Reduce body font from 20pt to 16pt across all slide templates.
Re-test alignment on all seven verification decks.
Keep heading at 22pt; the change is body text only.
Outcome: clean, consistent slides that render correctly on every machine.
E
Workstream E · Cost optimization & open stack
Lower per-deck cost + Travenza owns the code
Move inference to OpenRouter for per-node model selection, cut per-deck cost by 70–85%, and transfer repos, GCP project, secrets, and billing to Travenza accounts.
OpenRouter routing: eleven nodes ported, models switchable from a config table.
Quality gate: cutover happens only after a fixed sample matches today's quality bar.
Ownership handoff: repositories, Cloud Run, secrets, and billing transferred with a runbook.
Post-handoff cover: cost and failure-rate monitoring; model retuning if needed.
Out of scope
New PPT templates, redesign of slide layouts, or new product features.
Multi-tenant SaaS packaging or white-label sales channels.
ERP integration changes or new data-source connectors.
Full rebuild or re-platforming of the generation pipeline.
Already in Motion
~1 month of work delivered
The stabilization engagement — delivered at a fixed fee of ₹20,000 — closed the reliability gap. Image formats were fixed so downloads keep their pictures, generation was hardened so the whole ops team can produce decks, and layout updates from the July review (summary length, fonts, cost styling, T&C) are live. Seven test decks were verified end to end. That work forms the baseline this proposal builds on.
In parallel, exploratory work on a new agentic harness has produced early results that inform the image-relevance and description-quality workstreams above. That R&D is not billed here and is offered as a foundation for faster, better execution on items A and C.
The delivery report documenting all stabilization changes is available separately.
What it does
Broadens image sourcing and relevance so decks reflect the actual itinerary.
Adds itinerary history with prompt persistence — never lose a pasted input.
Expands day-wise descriptions into richer, more vivid copy per day slide.
Reduces body font from 20pt to 16pt, restoring slide alignment.
Routes all generation through OpenRouter with per-node model selection, targeting 70–85% cost reduction.
Transfers repositories, Cloud Run, secrets, and billing to Travenza accounts with a deploy runbook.
Build Timeline
One month of shipping, one month of cover
All five workstreams delivered within one month. Month 2 is monitoring and any cleanup.
Image relevance + font fix + history
Week 1–2
Broaden image providers with relevance-weighted queries; reduce body font to 16pt and verify alignment; build input persistence and history view.
Day descriptions + OpenRouter routing
Week 2–3
Expand day-wise generation prompts; wire OpenRouter, port eleven node calls, run quality-gate pilot on the fixed sample set.
Cutover + ownership handoff
Week 3–4
Quality gate pass → cutover to OpenRouter; transfer repos, GCP project, secrets, and billing to Travenza accounts; write deploy runbook.
Operating cover
Month 2
Monitor cost and failure rate; fix any migration issues; retune steps that underperform on real decks.
Cost
Fixed retainer, prepaid per month
A 2-month retainer covering all five workstreams. Payment is prepaid at the start of each month. Scope changes (new features, redesign, ERP work) are quoted separately.
Month 2Cutover + ownership handoff · Monitoring · Model retuning · Remaining polish on A–D₹25,000
Payment50% upfront (₹25,000) to start · ₹25,000 at beginning of Month 2Prepaid per month
Total · 2 months₹50,000
What this retainer replaces (if bought separately)
Item
Standalone cost
Packaged here
Cost optimization + open stack (original proposal)
₹68,000
Included
Image quality improvements
Separately quoted
Included
Itinerary history + prompt persistence
Separately quoted
Included
Day-wise descriptions + font alignment
Separately quoted
Included
Total value
₹68,000+
₹50,000
The cost optimization workstream (E) alone was previously quoted at ₹68,000 over three months. This retainer delivers it alongside the four quality improvements in two months at a lower total. The team gets more, faster, for less — with the retainer structure keeping scope bounded and payment predictable.
Per-deck cost after optimization
Model cost todayMeasured production baseline~₹55 (~$0.58)
Target after OpenRouter switchPilot confirms the final figure~₹8–17 (~$0.08–0.18)
Reduction70–85%
Monthly API spend at actual volume (100+ decks/month)
Decks / month
Current (≈)
Target (≈)
Monthly savings (≈)
50
₹2,750
₹500
₹2,250
100
₹5,500
₹1,000
₹4,500
150
₹8,250
₹1,500
₹6,750
200
₹11,000
₹2,000
₹9,000
At 100 decks/month — the team's stated target — the monthly API saving alone is ~₹4,500. After the retainer ends, that saving stays permanently. At higher volumes the retainer cost is recovered through savings within a few quarters.
Business Outcome
What changes for Travenza
Images are relevant and destination-specific — not generic stock photos.
Every pasted itinerary is saved with its prompt — revisitable, re-generatable, trackable.
Day slides tell a story with richer activity descriptions tuned for client presentation.
Slides render cleanly on every machine with body text at 16pt.
aayush built the current pipeline (an eleven-node LangGraph system that reads a raw itinerary and produces a complete, branded PPT deck) and runs it in production on Cloud Run. The stabilization engagement delivered ~30 changes across API and frontend: image reliability, generation hardening, layout updates, and platform security — verified end to end with seven test decks. That baseline is the foundation for this retainer.
Conclusion
90% efficiency. Cheaper to run. Owned by Travenza.
Four team-identified gaps, one cost + ownership restructuring, two months. The stabilization got the tool back to baseline; this retainer closes the gap to 90% efficiency and puts the stack in Travenza's hands. ₹50,000 fixed — prepaid per month, with 50% upfront to start.