ProductLens — Interactive Demo

ProductLens is a lightweight Gradio demo that helps you compare consumer products/services based on your priorities (privacy, price, ease of setup, etc.). It runs a simple agent pipeline:

1. Planner → identifies products + evaluation criteria + search queries
2. Research → performs web searches and summarizes findings
3. Comparator → ranks options and produces a recommendation + comparison table

---

Architecture

User query → Planner Agent → Search Agent(s) (parallel) → Comparator Agent → UI

• planner_agent.py: turns your query into a structured plan (ProductPlan)
• search_agent.py: calls a web_search tool and summarizes results
• comparator_agent.py: synthesizes a decision + markdown table
• comparison_manager.py: orchestrates the workflow
• load_vars.py: centralized provider/model configuration

+----------------------------------------------------------------------------------+
|                               Hugging Face Space / Docker                        |
|                                                                                  |
|  +---------------------------+                                                   |
|  |      Gradio UI (7860)     |                                                   |
|  |  - Textbox + Examples     |                                                   |
|  |  - Streams status output  |                                                   |
|  +-------------+-------------+                                                   |
|                | user query                                                      |
|                v                                                                 |
|  +---------------------------+        +---------------------------------------+  |
|  |    COMPARISON MANAGER      |       |              Env Config               |  |
|  |  - trace_id + orchestration|<------|  PROVIDER + MODEL + KEYS + N settings |  |
|  |  - async generator output  |       +---------------------------------------+  |
|  +-------------+-------------+                                                   |
|                |                                                                 |
|                | (1) PLAN: ProductPlan (JSON schema)                             |
|                v                                                                 |
|  +---------------------------+                                                   |
|  |       PLANNER AGENT       |                                                   |
|  |  - pick N products        |                                                   |
|  |  - derive criteria        |                                                   |
|  |  - generate searches      |                                                   |
|  +-------------+-------------+                                                   |
|                | Runner.run()                                                    |
|                v                                                                 |
|  +---------------------------+                                                   |
|  | LLM PROVIDER ROUTER       |                                                   |
|  |  OpenAI / OpenRouter /    |                                                   |
|  |  Ollama (ChatCompletions) |                                                   |
|  +---------------------------+                                                   |
|                |                                                                 |
|                | (2) RESEARCH: parallel per-product                              |
|                v                                                                 |
|  +-----------------------------------------+                                     |
|  |  Async Task Fan-out (asyncio.gather)    |                                     |
|  |  - one task per product                 |                                     |
|  +-------------------+---------------------+                                     |
|                      |                                                           |
|          +-----------+---------------+-----------------------+----   ...         |
|          |                           |                       |                   |
|          v                           v                       v                   |
|  +--------------------+   +--------------------+                                 |
|  |    SEARCH AGENT    |   |    SEARCH AGENT    |            ...                  |
|  |  tool_choice=req'd |   |  tool_choice=req'd |                                 |
|  +---------+----------+   +---------+----------+                                 |
|            | tool call            | tool call                                    |
|            v                      v                                              |
|     +-------------+        +-------------+                                       |
|     | web_search  |        | web_search  |  (function_tool)                      |
|     | (Ollama API)|        | (Ollama API)|                                       |
|     +------+------+        +------+------+                                       |
|            |                      |                                              |
|            v                      v                                              |
|     +-------------+        +-------------+                                       |
|     | Web results |        | Web results |                                       |
|     +------+------+        +------+------+                                       |
|            | summary              | summary                                      |
|            +-----------+----------+                                              |
|                        v                                                         |
|          +---------------------------+                                           |
|          |    research_results[]     |                                           |
|          |    "product: summary"     |                                           |
|          +-------------+-------------+                                           |
|                        |                                                         |
|                        | (3) COMPARE: decision + table                           |
|                        v                                                         |
|          +---------------------------+                                           |
|          |     COMPARATOR AGENT      |                                           |
|          |  - normalize differences  |                                           |
|          |  - rank options           |                                           |
|          |  - output markdown table  |                                           |
|          +-------------+-------------+                                           |
|                        | Runner.run()                                            |
|                        v                                                         |
|          +---------------------------+                                           |
|          | Final Output -> Gradio UI |                                           |
|          | - Recommendation          |                                           |
|          | - Comparison Table        |                                           |
|          +---------------------------+                                           |
|                                                                                  |
+----------------------------------------------------------------------------------+

---

Notes

• Providers/models are selected via PROVIDER and *_MODEL env vars.
• If you use free-tier OpenRouter models, you may hit rate limits or temporary throttling.
• Increasing NUM_OF_PRODUCTS (default=2) and NUM_OF_SEARCHES (default=1) increases cost/latency and may exceed small context windows.

---

Usage

Try prompts like:

• “Best noise-canceling headphones for travel and calls”
• “Privacy-focused smart doorbells”
• “Robot vacuum for pet hair, quiet, easy maintenance”