Agent Requirements

Agent Objective:

What is the agent designed to accomplish? Define the core task and success criteria...

1. Provide answer to user's question
2. retrieve relevant document
3. answer should have source citation

Autonomy & Constraints:

What level of autonomy does the agent have? What are the boundaries and constraints it must operate within?

1. chunking of document while saving
2. re-ranking
3. hallucination detection
4. handling question span across multiple documents

Tool & Action Design

Define the tools and actions available to the agent. Specify tool interfaces, input/output schemas, and how the agent selects which tool to use...

1. search user's query

Agent Architecture

Design the overall agent architecture. Choose an orchestration pattern (single agent, multi-agent, hierarchical). Identify key components and how they interact. Use the diagramming tool to illustrate the architecture.

i will go with single agent architecture

Flow -1 - ingestion flow

1. Documents from source will be chunked and by using an textEmbedding generation model, its embeddings will be generated.
2. Store the embeddings in a vector db
3. in case of the query from user, the coresponding embeddings will be generated for the query
4. the embeddings which have closest cosine similarity with the query embedding, will be picked as highest score and retrieved .
5. To avoid the hallucination of the agent, while retrieval, we can set the model temperture close to 0 , so that model's response will be more determinastic

Memory & Context Design

How does the agent manage memory and context? Define short-term (conversation) and long-term (persistent) memory strategies. How is context window managed? What retrieval mechanisms are used (RAG, vector stores, summarization)?

For keep track of memory, user's query will be appended with AI response and sent to LLM on every request.

Once the conversation reaches the context window max size, it will be summarized and stored as session with a sessionid

session will be in the ephemoral memory,

Core Reasoning

Deep dive into the agent's reasoning strategy. How does it plan, decide, and self-correct? Discuss prompt engineering, chain-of-thought, reflection loops, failure handling, and tradeoffs.

based on the closest cosine similarity from vector db , record will be fetched. The record having highest similarity score will be sent to LLM for response generation.

for failure handling, i can implement the retry with Exponential backoff mechanism. So that if any call fails, before failing to user, request will be retried 3 times with an expontial delay

Safety & Permissions

What guardrails and safety mechanisms does the agent have? Define permission boundaries, human-in-the-loop checkpoints, content filtering, rate limiting, and how the agent handles adversarial inputs or unexpected states.

guardRails

• while talking with the agent, agent should not entertain any kind of Derogatory and prohibited words from user.
• To identify these words, an appropriate prompt should be provided to the model.
• Model should not process any PII data
• In case of any stage decision making is needed, agent should ask for human approval