Recall and Commit

Inspiration

We often miss the small commitments we give when we are busy with other works. So, we need a thing which acts as a remainder to list all the commitments automatically for us based on our conversations.

What it does

It listens to all the conversations of that user and analyzes them using LLM's and check if it is a commitment and if yes it displays it to the user. Also a recall feature can used to go ask AI why I have taken that decision in the past. It analyzes the conversations and provide the answer.

How we built it

I built it using React for Front end and Fast API for backend. I used Langchain and LangGraph for AI Agents and Gemini 3.8-flash as the LLM decision maker.

Challenges we ran into

Telling real commitments from casual talk: "I'll try to" and "maybe I can" are very different from "I will send it by Friday." Tuning the prompts to cut false positives took many iterations.

Vague deadlines: Phrases like "next week" or "after the meeting" had to be turned into concrete dates using the conversation's timestamp.

Context across messages: A commitment often depends on earlier messages, so a single message isn't enough to judge it.

Recall accuracy: I had to make sure answers came from what the user actually said and not from the model guessing.

Accomplishments that we're proud of

I built a working end-to-end pipeline, from raw conversation to a tracked commitment, in a short time. Recall gives answers grounded in the user's own history, not generic AI responses.

What we learned

Good prompt design and structured outputs matter more than model size for classification tasks. LangGraph makes multi-step agent workflows much easier to debug and control than chaining calls by hand. Grounding answers in retrieved context is essential to prevent hallucinations in a memory-style product.

What's next for Recall and Commit

Next, I plan to integrate with Slack, Gmail, Microsoft Teams, and meeting transcripts so commitments are captured wherever they happen. I will also add smart reminders and calendar sync, with nudges before a deadline. Finally, I plan to integrate with Alexa+ once the developer tools become available.

Built With

• fastapi
• langchain
• langgraph
• python
• react