AI Travel Planner - March 27th, 2025 - Tripmate - PLDG 2

[Dhruv-Varshney-developer]

AI Travel Planner - March 27th, 2025 - Tripmate - PLDG 2

Contributors

• @Dhruv-Varshney-developer
• @Nkovaturient

Accomplished Tasks

• Uploading Trip Data to Storacha decentralized storage (Agent 1)
• Integrating Telegram bot conversational history storage
• Switching from Vertex AI, GCP & SerpAPI to the direct Gemini API for Agent 2
• Implementing robust caching and code refactoring in Agent 2
• Successfully fetching flight, hotel, and attraction details in Agent 2

Update

This week was a rollercoaster of technical challenges and breakthroughs! We made significant progress on our AI Travel Planner, focusing on two key areas:

1. Uploading Trip Data to Storacha

Our primary task was to store conversational trip-related information from users interacting with the Telegram bot on the Storacha decentralized storage network.

Workflow Breakdown:

Agent TripMate interacts with users → Conversational history stored on sqlite3 → As the agent interacts with users, the convo/trip data is concurrently stored via storeTripData actions → Agent A receives the requisite details, then uploads them on Storacha → Retrieves/returns (STORAGE_RETRIEVE) the CID to users (IPFS Gateway URL).

Key Achievements:

• Storage client is initialized via storagePlugin.
• Fixed CBOR Decode err caused by malfunctioned Agent delegation.
• Successfully accessed and extracted the conversational history through registered actions storeTripData and retrieveTripData on runtime agent using recentMessagesData object, converting it into a string, blob-like, then a file to be uploaded on Storacha via uploadFile.
• Ensured that while interacting with users, our agent triggers storeTripData to save travel info.
• Fixed the failed space/blob/add invocation error caused by a delegation issue by recreating the Agent.
• Successfully uploaded data to Storacha and accessed it via the IPFS HTTP gateway URL.

2. Agent 2 Improvements:

• Switched to direct Gemini API for simplified authentication.
• Fixed flight and hotel search parsing by correcting key-value pairs.
• Implemented caching to prevent redundant API calls.
• Modularized codebase into specialized services.
• Enhanced error handling and user experience.
• Removed attraction API calls by leveraging Gemini’s knowledge base.
• Added more flexibility in search parameters.
• Integrated airport code mappings.

Demo Videos:

1. Trip Data Upload to Storacha
   
2. Tripmate Agent-2 in Action

Tripmate.-.Agent.2.-.26th.march.mov

---

Questions

We have a few questions where we’d need the team’s guidance, suggestions and opinions:

1. Regarding Lit Encryption:

Should we encrypt the data before uploading?
If yes, then we need to use the sqlite3 DB that is currently storing the conversational history locally for Agent A to familiarize itself with users’ past preferences and interests.

Workflow if we keep SQLite3:

Agent TripMate interacts with users → Conversational history stored on sqlite3 → As the agent interacts with users, it is concurrently stored via storeTripData actions → Agent A receives the requisite details, then uploads it on Storacha → Retrieves/returns (STORAGE_RETRIEVE) the CID to users (IPFS Gateway URL).

If we remove SQLite3, we need to reconfigure our AgentRuntime with databaseAdapter type DB in index.ts.

2. User Search Data Extraction:

• What data points should we extract from a user’s search to improve trip planning?
• Currently, we are targeting flights, hotels, and attractions as a proof of concept. Would love any suggestions on additional factors we should consider.

3. Alternative APIs for Train Data:

• SerpAPI does not support train data. Does anyone have recommendations for a reliable API for train bookings? As I searched - I realized there are country-specific train data APIs and some APIs provide data for multiple countries. But is there an API that provide all at one place?

---

Potential Bottlenecks & API Costs

One issue we might face is exhausting free quotas on APIs such as:

• SerpAPI (for real-time travel data)
• Gemini (Agent 2)
• OpenRouter (Agent 1)

Funding Options:

• Does anyone know of grants, resources, or credits that could help cover these costs?
• While we can build the PoC without additional resources, we could hit limitations with just 10 users, making even early adoption challenging.

---

Next Tasks

• Review and merge existing pull requests.
• Implement Storacha retrieval mechanism in Agent 2.
• Connect agents and Sync agents properly.
• Explore and implement data encryption options.
• Evaluate comprehensive alternative API solutions and data option.
• Deploy the MVP for testing.

---

Links:

Reference PRs for this week:

1. PR #2 - Lit Storacha Agents
2. PR #2 - TripMate
3. Work for First Agent
4. Second Agent (TripMate)

[jamiechicago312]

Retrieve Data from Storacha:

• Check this out https://docs.storacha.network/how-to/retrieve/ to see how to retrieve the data

Lit Encryption

• Client worked on for encrypting data & storing on LIT
• Our repo we are working on https://github.com/storacha/lit-storacha-demo
• Shared by @seetadev : https://drive.google.com/drive/u/5/folders/1dbxjNMvvJusAuUTA9TietqsXJnI9QiTG?authuser=0

[Nkovaturient]

This would be of great help!
Thanks a lot @jamiechicago312 for sharing!

[alanshaw]

The lit encryption client will be released when this PR merges - storacha/upload-service#207