Manual rebuild of the chain

notebooks/transcript_rag.ipynb

@@ -59,7 +59,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 5,
    "metadata": {},
    "outputs": [
     {
@@ -86,7 +86,7 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "2025-05-31 00:25:16,293 - INFO - <module> - Loaded .env file\n"
+      "2025-05-31 01:01:57,165 - INFO - <module> - Loaded .env file\n"
      ]
     }
    ],
@@ -103,23 +103,17 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "2025-05-31 00:25:18,096 - INFO - print - Configuration parameters:\n",
-      "2025-05-31 00:25:18,097 - INFO - print -   eva_workflow_name: EVA_workflow\n",
-      "2025-05-31 00:25:18,099 - INFO - print -   eva_log_level: INFO\n",
-      "2025-05-31 00:25:18,100 - INFO - print -   transcript_glob: ./data/test.json\n",
-      "2025-05-31 00:25:18,101 - INFO - print -   embedding_model: mxbai-embed-large\n",
-      "2025-05-31 00:25:18,102 - INFO - print -   eva_strip_think: True\n",
-      "2025-05-31 00:25:18,103 - INFO - print -   embedding_api: ModelAPI.OLLAMA\n",
-      "2025-05-31 00:25:18,104 - INFO - print -   llm_api: ModelAPI.OLLAMA\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "2025-05-31 00:25:18,104 - INFO - print -   max_research_loops: 2\n",
-      "2025-05-31 00:25:18,105 - INFO - print -   llm_tool_model: qwen3:4b\n",
-      "2025-05-31 00:25:18,106 - INFO - print -   n_context_docs: 3\n"
+      "2025-05-31 01:01:59,036 - INFO - print - Configuration parameters:\n",
+      "2025-05-31 01:01:59,037 - INFO - print -   eva_workflow_name: EVA_workflow\n",
+      "2025-05-31 01:01:59,038 - INFO - print -   eva_log_level: INFO\n",
+      "2025-05-31 01:01:59,039 - INFO - print -   transcript_glob: ./data/test.json\n",
+      "2025-05-31 01:01:59,040 - INFO - print -   embedding_model: text-embedding-3-small\n",
+      "2025-05-31 01:01:59,042 - INFO - print -   eva_strip_think: True\n",
+      "2025-05-31 01:01:59,042 - INFO - print -   embedding_api: ModelAPI.OPENAI\n",
+      "2025-05-31 01:01:59,042 - INFO - print -   llm_api: ModelAPI.OPENAI\n",
+      "2025-05-31 01:01:59,043 - INFO - print -   max_research_loops: 2\n",
+      "2025-05-31 01:01:59,043 - INFO - print -   llm_tool_model: gpt-4.1-mini\n",
+      "2025-05-31 01:01:59,044 - INFO - print -   n_context_docs: 3\n"
      ]
     }
    ],
@@ -136,8 +130,8 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "2025-05-31 00:25:19,818 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:19,844 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n"
+      "2025-05-31 01:01:59,776 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:00,223 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n"
      ]
     },
     {
@@ -166,22 +160,22 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "2025-05-31 00:25:20,879 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:21,701 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:22,528 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:23,356 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:23,851 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:24,806 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:25,058 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:25,730 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:26,550 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:27,421 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:28,372 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:28,708 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:29,072 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:29,335 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:29,700 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:30,011 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n"
+      "2025-05-31 01:02:00,707 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:00,717 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:00,786 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:00,830 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:01,060 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:01,092 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:01,234 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:01,452 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:01,543 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:01,773 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:02,459 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:02,951 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:03,020 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:03,075 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:03,301 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:05,029 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n"
      ]
     },
     {
@@ -217,8 +211,8 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "2025-05-31 00:25:30,144 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/embed \"HTTP/1.1 200 OK\"\n",
-      "2025-05-31 00:25:40,574 - INFO - _send_single_request - HTTP Request: POST http://127.0.0.1:11434/api/chat \"HTTP/1.1 200 OK\"\n"
+      "2025-05-31 01:02:05,322 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/embeddings \"HTTP/1.1 200 OK\"\n",
+      "2025-05-31 01:02:08,523 - INFO - _send_single_request - HTTP Request: POST https://api.openai.com/v1/chat/completions \"HTTP/1.1 200 OK\"\n"
      ]
     }
    ],
@@ -237,19 +231,7 @@
      "text": [
       "==================================\u001b[1m Ai Message \u001b[0m==================================\n",
       "\n",
-      "\n",
-      "\n",
-      "A **layer** in Photoshop is a separate component of an image, acting like a \"flat pint of glass\" stacked on top of others. Each layer holds its own content (like a photo or artwork) and can be hidden or shown using the **Eye icon** in the Layers panel.  \n",
-      "\n",
-      "Key points from the video:  \n",
-      "- Layers are the building blocks of any image in Photoshop.  \n",
-      "- They allow you to edit parts of an image independently without affecting other parts.  \n",
-      "- Transparent areas in a layer let you see through to layers below (e.g., a checkerboard pattern in the example).  \n",
-      "- The **Layers panel** lets you manage and select individual layers for editing.  \n",
-      "\n",
-      "Timestamp: **00:00:00–00:00:23** (first explanation).  \n",
-      "\n",
-      "This is covered in the \"Understand layers\" video. 🎨\n",
+      "A layer in Photoshop CC is like a separate flat pane of glass stacked on top of others. Each layer contains separate pieces of content that you can work on independently. Layers may have transparent areas that let you see through to layers below. The biggest benefit of using layers is being able to edit parts of an image independently without affecting the rest of the image (see 0:47–3:41 and 85:45–87:10). 🎨🖼️\n",
       "**REFERENCES**\n",
       "[\n",
       "  {\n",
@@ -261,14 +243,14 @@
       "  {\n",
       "    \"title\": \"Understand layers\",\n",
       "    \"source\": \"https://images-tv.adobe.com/avp/vr/b758b4c4-2a74-41f4-8e67-e2f2eab83c6a/f810fc5b-2b04-4e23-8fa4-5c532e7de6f8/e268fe4d-e5c7-415c-9f5c-d34d024b14d8_20170727011753.1280x720at2400_h264.mp4\",\n",
-      "    \"start\": 280.4,\n",
-      "    \"stop\": 284.58\n",
+      "    \"start\": 85.75,\n",
+      "    \"stop\": 152.97\n",
       "  },\n",
       "  {\n",
-      "    \"title\": \"Understand layers\",\n",
-      "    \"source\": \"https://images-tv.adobe.com/avp/vr/b758b4c4-2a74-41f4-8e67-e2f2eab83c6a/f810fc5b-2b04-4e23-8fa4-5c532e7de6f8/e268fe4d-e5c7-415c-9f5c-d34d024b14d8_20170727011753.1280x720at2400_h264.mp4\",\n",
-      "    \"start\": 85.75,\n",
-      "    \"stop\": 279.91\n",
+      "    \"title\": \"Get organized with layer groups\",\n",
+      "    \"source\": \"https://videos-tv.adobe.com/2013-07-23/f65b5a0ef188ba5e5a96df93a8ead3cf.mp4\",\n",
+      "    \"start\": 181.35,\n",
+      "    \"stop\": 239.72\n",
       "  }\n",
       "]\n"
      ]
@@ -286,9 +268,9 @@
     {
      "data": {
       "text/plain": [
-       "[Document(metadata={'video_id': 19172, 'title': 'Understand layers', 'desc': 'Learn what layers are and why they are so useful.', 'length': '00:04:44.75', 'group': 'data/test.json', 'source': 'https://images-tv.adobe.com/avp/vr/b758b4c4-2a74-41f4-8e67-e2f2eab83c6a/f810fc5b-2b04-4e23-8fa4-5c532e7de6f8/e268fe4d-e5c7-415c-9f5c-d34d024b14d8_20170727011753.1280x720at2400_h264.mp4', 'speech_start_stop_times': [[0.47, 3.41], [3.81, 9.13], [9.309999, 15.01], [15.299999, 20.57], [20.88, 23.3], [23.83, 27.93], [29.38, 32.79], [32.96, 33.92], [34.43, 40.21], [41.91, 45.37], [45.88, 49.01], [49.54, 55.130001], [55.72, 58.49], [58.72, 62.14]], 'start': 0.47, 'stop': 62.14, '_id': 21, '_collection_name': 'a31ff6bb-4d35-4b52-8612-55df01f799fa'}, page_content=\"Layers are the building blocks of any image in Photoshop CC. So, it's important to understand, what layers are and why to use them - which we'll cover in this video. If you're following along, open this layered image from the downloadable practice files for this tutorial. You might think of layers like separate flat pints of glass, stacked one on top of the other. Each layer contains separate pieces of content. To get a sense of how layers are constructed, let's take a look at this Layers panel. I've closed my other panels, so that we can focus on the Layers panel. But you can skip that. By the way: If your Layers panel isn't showing, go up to the Window menu and choose Layers from there. The Layers panel is where you go to select and work with layers. In this image there are 4 layers, each with separate content. If you click the Eye icon to the left of a layer, you can toggle the visibility of that layer off and on. So, I'm going to turn off the visibility of the tailor layer. And keep your eye on the image, so you can see what's on that layer.\"),\n",
-       " Document(metadata={'video_id': 19172, 'title': 'Understand layers', 'desc': 'Learn what layers are and why they are so useful.', 'length': '00:04:44.75', 'group': 'data/test.json', 'source': 'https://images-tv.adobe.com/avp/vr/b758b4c4-2a74-41f4-8e67-e2f2eab83c6a/f810fc5b-2b04-4e23-8fa4-5c532e7de6f8/e268fe4d-e5c7-415c-9f5c-d34d024b14d8_20170727011753.1280x720at2400_h264.mp4', 'speech_start_stop_times': [[280.4, 284.58]], 'start': 280.4, 'stop': 284.58, '_id': 24, '_collection_name': 'a31ff6bb-4d35-4b52-8612-55df01f799fa'}, page_content=\"There's lots more to learn about layers, so stay tuned for the rest of this tutorial.\"),\n",
-       " Document(metadata={'video_id': 19172, 'title': 'Understand layers', 'desc': 'Learn what layers are and why they are so useful.', 'length': '00:04:44.75', 'group': 'data/test.json', 'source': 'https://images-tv.adobe.com/avp/vr/b758b4c4-2a74-41f4-8e67-e2f2eab83c6a/f810fc5b-2b04-4e23-8fa4-5c532e7de6f8/e268fe4d-e5c7-415c-9f5c-d34d024b14d8_20170727011753.1280x720at2400_h264.mp4', 'speech_start_stop_times': [[85.75, 88.659999], [89.42, 100.11], [101.469999, 108.64], [109.09, 117.459999], [117.75, 129.45], [129.97, 133.37], [133.73, 143.98], [144.76, 152.97], [153.7, 155.87], [156.139999, 161.27], [161.789999, 165.03], [165.31, 170.53], [171.059999, 175.409999], [175.59, 186.54], [187.29, 189.44], [189.679999, 199.77], [200.07, 205.68], [205.89, 208.47], [208.76, 211.96], [212.25, 213.87], [214.289999, 222.08], [222.12, 227.62], [228.33, 233.199999], [233.57, 236.77], [237.25, 244.679999], [245.08, 255.14], [255.41, 261.84], [262.219999, 273.1], [273.32, 279.91]], 'start': 85.75, 'stop': 279.91, '_id': 23, '_collection_name': 'a31ff6bb-4d35-4b52-8612-55df01f799fa'}, page_content=\"Now let's take a look at just one layer, the tailor layer. A quick way to turn off all the layers except the tailor layer, is to hold down the Option key on the Mac, or the ALT key on the PC, and click on the Eye icon to the left of the tailor layer. In the Document window, you can see that this layer contains just the one small photo surrounded by a gray and white checkerboard pattern. That pattern represents transparent pixels, which allow us to see down through the corresponding part of this layer to the content of the layers below. So, let's turn that content back on by going back to the Layers panel, again holding the Option key on the Mac or the ALT key on the PC and clicking on the Eye icon to the left of the tailor layer. And all the other layers and their Eye icons come back into view. So again: You might think of layers like a stack of pints of glass, each with its own artwork and in some cases transparent areas that let you see down through to the layers below. The biggest benefit of having items on separate layers like this, is that you'll be able to edit pieces of an image independently without affecting the rest of the image. Follow along with this example to see what I mean. Let's say you wanted to edit or move just the small photo of the cloth being cut. To do that, you'd have to select the layer that contains that photo. Go to the Layers panel and click on the cloth layer just to the right of the layer name. Now that the cloth layer is selected, any editing you do will affect only that layer. For example, go to the Tools panel and select the Move tool and then click on the cloth photo and drag - and only that photo moves, because only that layer is selected. And this applies to other kinds of edits too. For example, if you painted on the image or applied a filter, as you'll do later in this tutorial series, those edits would affect only the selected cloth layer and not the rest of the image. By the way: There are some editing activities, like moving layers, that you can do to more than one layer at a time. And to do that, you need to select more than one layer. Over in the Layers panel, we already have the cloth layer selected. If you don't, click on that layer. To select the tailor layer too, hold down the Command key on a Mac or the Ctrl key on a PC, and click on the tailor layer. With both layers selected, you can click and drag either of those photos and they'll move together. Now here's a bonus tip: With the Move tool selected, take a look at its Options bar. And you'll see this option, Auto-Select: Layer is checked. With that option checked, clicking anywhere in the image will automatically select the top layer at that location that contains content. For example, watch the Layers panel as I click over here on the pattern photo and you'll see that the pattern layer was automatically selected for you in the Layers panel. Sometimes that's really useful, but sometimes it can cause you to inadvertently select a layer that you didn't mean to. To avoid that you can uncheck Auto-Select: Layer in the Move tool Options bar and instead select layers manually by clicking on them in the Layers panel, as we did in this tutorial. So, to recap: Being able to work with selected pieces of content independently is the biggest benefit of using layers.\")]"
+       "[Document(metadata={'video_id': 19172, 'title': 'Understand layers', 'desc': 'Learn what layers are and why they are so useful.', 'length': '00:04:44.75', 'group': 'data/test.json', 'source': 'https://images-tv.adobe.com/avp/vr/b758b4c4-2a74-41f4-8e67-e2f2eab83c6a/f810fc5b-2b04-4e23-8fa4-5c532e7de6f8/e268fe4d-e5c7-415c-9f5c-d34d024b14d8_20170727011753.1280x720at2400_h264.mp4', 'speech_start_stop_times': [[0.47, 3.41], [3.81, 9.13], [9.309999, 15.01], [15.299999, 20.57], [20.88, 23.3], [23.83, 27.93], [29.38, 32.79], [32.96, 33.92], [34.43, 40.21], [41.91, 45.37], [45.88, 49.01], [49.54, 55.130001], [55.72, 58.49], [58.72, 62.14]], 'start': 0.47, 'stop': 62.14, '_id': 21, '_collection_name': 'a19921d4-b5da-4707-bf34-1b1e3828d581'}, page_content=\"Layers are the building blocks of any image in Photoshop CC. So, it's important to understand, what layers are and why to use them - which we'll cover in this video. If you're following along, open this layered image from the downloadable practice files for this tutorial. You might think of layers like separate flat pints of glass, stacked one on top of the other. Each layer contains separate pieces of content. To get a sense of how layers are constructed, let's take a look at this Layers panel. I've closed my other panels, so that we can focus on the Layers panel. But you can skip that. By the way: If your Layers panel isn't showing, go up to the Window menu and choose Layers from there. The Layers panel is where you go to select and work with layers. In this image there are 4 layers, each with separate content. If you click the Eye icon to the left of a layer, you can toggle the visibility of that layer off and on. So, I'm going to turn off the visibility of the tailor layer. And keep your eye on the image, so you can see what's on that layer.\"),\n",
+       " Document(metadata={'video_id': 19172, 'title': 'Understand layers', 'desc': 'Learn what layers are and why they are so useful.', 'length': '00:04:44.75', 'group': 'data/test.json', 'source': 'https://images-tv.adobe.com/avp/vr/b758b4c4-2a74-41f4-8e67-e2f2eab83c6a/f810fc5b-2b04-4e23-8fa4-5c532e7de6f8/e268fe4d-e5c7-415c-9f5c-d34d024b14d8_20170727011753.1280x720at2400_h264.mp4', 'speech_start_stop_times': [[85.75, 88.659999], [89.42, 100.11], [101.469999, 108.64], [109.09, 117.459999], [117.75, 129.45], [129.97, 133.37], [133.73, 143.98], [144.76, 152.97]], 'start': 85.75, 'stop': 152.97, '_id': 23, '_collection_name': 'a19921d4-b5da-4707-bf34-1b1e3828d581'}, page_content=\"Now let's take a look at just one layer, the tailor layer. A quick way to turn off all the layers except the tailor layer, is to hold down the Option key on the Mac, or the ALT key on the PC, and click on the Eye icon to the left of the tailor layer. In the Document window, you can see that this layer contains just the one small photo surrounded by a gray and white checkerboard pattern. That pattern represents transparent pixels, which allow us to see down through the corresponding part of this layer to the content of the layers below. So, let's turn that content back on by going back to the Layers panel, again holding the Option key on the Mac or the ALT key on the PC and clicking on the Eye icon to the left of the tailor layer. And all the other layers and their Eye icons come back into view. So again: You might think of layers like a stack of pints of glass, each with its own artwork and in some cases transparent areas that let you see down through to the layers below. The biggest benefit of having items on separate layers like this, is that you'll be able to edit pieces of an image independently without affecting the rest of the image.\"),\n",
+       " Document(metadata={'video_id': 4157, 'title': 'Get organized with layer groups', 'desc': 'Learn some great tips for working with layers.', 'length': '00:04:05.78', 'group': 'data/test.json', 'source': 'https://videos-tv.adobe.com/2013-07-23/f65b5a0ef188ba5e5a96df93a8ead3cf.mp4', 'speech_start_stop_times': [[181.35, 182.05], [183.16, 184.01], [184.44, 186.49], [186.62, 187.46], [187.97, 191.469999], [191.79, 194.43], [195.0, 203.05], [203.13, 205.93], [206.37, 209.55], [210.09, 212.449999], [212.46, 215.060001], [215.609999, 219.54], [220.02, 223.97], [225.37, 227.109999], [227.929999, 228.57], [229.07, 239.72]], 'start': 181.35, 'stop': 239.72, '_id': 3, '_collection_name': 'a19921d4-b5da-4707-bf34-1b1e3828d581'}, page_content='Click \"OK\". You did it again. Let\\'s talk about one more way to do this. Using searching. Let\\'s say we have a complicated document that has a lot of stuff in it. We have a lot of type or text layers. That\\'s a unique type of layer in the Layers panel, which got a bunch of them you\\'ve got a couple of hundred layers and these are scattered all over the place. And you want all of your type layers in one group. So we go to the Search facility here and click \"Kind\". And then click this button for just type layers. It isolates everything but those two. We select them with a \"Shift\" click, go up here, this is the easiest way to do it. Go into \"New Group from Layers\" and call it \"Text\" - type whatever you want. And then turn this back off again - here. And there you go. Creating groups gives you control, it reduces the chaos and clutter in the Layers panel and if at any time you change your mind, it\\'s a very easy thing to take them back out again.')]"
       ]
      },
      "execution_count": 13,
@@ -308,7 +290,7 @@
     {
      "data": {
       "text/plain": [
-       "Configuration(eva_workflow_name='EVA_workflow', eva_log_level='INFO', transcript_glob='./data/test.json', embedding_model='mxbai-embed-large', eva_strip_think=True, embedding_api=<ModelAPI.OLLAMA: 'OLLAMA'>, llm_api=<ModelAPI.OLLAMA: 'OLLAMA'>, max_research_loops=2, llm_tool_model='qwen3:4b', n_context_docs=3)"
+       "Configuration(eva_workflow_name='EVA_workflow', eva_log_level='INFO', transcript_glob='./data/test.json', embedding_model='text-embedding-3-small', eva_strip_think=True, embedding_api=<ModelAPI.OPENAI: 'OPENAI'>, llm_api=<ModelAPI.OPENAI: 'OPENAI'>, max_research_loops=2, llm_tool_model='gpt-4.1-mini', n_context_docs=3)"
       ]
      },
      "execution_count": 14,

pstuts_rag/pstuts_rag/graph.py

@@ -8,7 +8,8 @@ from langchain_core.messages import AIMessage
 from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
 from langchain_core.runnables import Runnable, RunnableLambda
 
-from langgraph.graph import StateGraph
+from langchain_openai import ChatOpenAI
+from langgraph.graph import StateGraph, START, END
 
 from pstuts_rag.prompts import SUPERVISOR_SYSTEM, TAVILY_SYSTEM
 from pstuts_rag.state import PsTutsTeamState
@@ -21,6 +22,7 @@ import logging
 from typing import Callable, Dict, Tuple, Optional, Union
 
 from langchain_huggingface import HuggingFaceEmbeddings
+from pstuts_rag.utils import ChatAPISelector
 
 from app import (
     ADOBEHELP,
@@ -116,8 +118,7 @@ def create_agent(
 
 
 def create_tavily_node(
-    llm: BaseChatModel, name: str = "AdobeHelp"
-) -> Tuple[Callable, AgentExecutor, TavilySearchResults]:
+     name: str = "AdobeHelp", config: Configuration = Configuration() ) -> Callable
     """Initialize tool, agent, and node for Tavily search of helpx.adobe.com.
 
     This function sets up a search agent that can query Adobe Photoshop help topics
@@ -133,6 +134,10 @@ def create_tavily_node(
             - The configured agent executor
             - The Tavily search tool instance
     """
+    
+    cls = ChatAPISelector.get(config.llm_api, ChatOpenAI)
+    llm = cls(model=config.llm_tool_model)
+
 
     adobe_help_search = TavilySearchResults(
         max_results=5, include_domains=["helpx.adobe.com"]
@@ -147,7 +152,11 @@ def create_tavily_node(
     return adobe_help_node
 
 
-def create_team_supervisor(llm: BaseChatModel, system_prompt, members):
+def create_team_supervisor(
+    system_prompt,
+    members,
+    config: Configuration = Configuration(),
+):
     """An LLM-based router."""
     options = ["FINISH"] + members
     function_def = {
@@ -167,6 +176,10 @@ def create_team_supervisor(llm: BaseChatModel, system_prompt, members):
             "required": ["next"],
         },
     }
+
+    cls = ChatAPISelector.get(config.llm_api, ChatOpenAI)
+    llm = cls(model=config.llm_tool_model)
+
     prompt = ChatPromptTemplate.from_messages(
         [
             ("system", system_prompt),
@@ -176,60 +189,27 @@ def create_team_supervisor(llm: BaseChatModel, system_prompt, members):
     ).partial(options=str(options), team_members=", ".join(members))
     return (
         prompt
-        | llm.bind_functions(functions=[function_def], function_call="route")
+        | llm.bind_tools(
+            tools=[function_def],
+            tool_choice={"type": "function", "function": {"name": "route"}},
+        )
         | JsonOutputFunctionsParser()
     )
 
 
-async def startup(
-    config=Configuration(), on_loading_complete: Optional[Callable] = None
-):
-    """
-    Initialize the application with optional loading completion callback.
-
-    Args:
-        config: Configuration object with application settings
-        on_loading_complete: Optional callback (sync or async) to call when
-                           datastore loading completes
-
-    Returns:
-        DatastoreManager: The initialized datastore manager
-    """
-
-    ### PROCESS THE CONFIGURATION
-    log_level = getattr(logging, config.eva_log_level, logging.INFO)
-    logging.basicConfig(level=log_level, format="%(levelname)s: %(message)s")
-
-    ### CREATE THE DATABASE
+def initialize_datastore(callback: Optional[Callable] = None):
 
-    datastore = DatastoreManager(
-        name=config.eva_workflow_name,
-        embeddings=HuggingFaceEmbeddings(model_name=config.embedding_model),
+    datastore = DatastoreManager()
+    if callback:
+        datastore.add_completion_callback(callback)
+    asyncio.create_task(
+        datastore.from_json_globs(Configuration().transcript_glob)
     )
 
-    ### START DATABASE POPULATION
-
-    globs = [str(g) for g in config.transcript_glob.split(":")]
-
-    # Add custom callback if provided, otherwise use default logging
-    if on_loading_complete:
-        datastore.add_completion_callback(on_loading_complete)
-    else:
-        # Default callback for logging
-        def default_logging_callback():
-            logging.info("🎉 Datastore loading completed!")
-
-        datastore.add_completion_callback(default_logging_callback)
-
-    asyncio.create_task(datastore.from_json_globs(globs))
-
-    ### CREATE THE RAG CHAIN
-    ai_graph = StateGraph(PsTutsTeamState, config_schema=Configuration)
-
     return datastore
 
 
-async def build_the_graph(current_state: ApplicationState):
+async def build_the_graph(datastore: DatastoreManager, config:Configuration=Configuration()):
     """
     Builds the agent graph for routing user queries.
 
@@ -239,33 +219,24 @@ async def build_the_graph(current_state: ApplicationState):
     Args:
         current_state: Current application state with required components
     """
-    adobe_help_node, _, _ = create_tavily_node(
-        llm=app_state.llm, name=ADOBEHELP
-    )
 
-    rag_node, _ = create_rag_node(
-        rag_chain=create_transcript_rag_chain(),
-        name=VIDEOARCHIVE,
-    )
+    adobe_help_node = create_tavily_node(name=ADOBEHELP, config=config)
+
+    rag_node = create_transcript_rag_chain(datastore, config=config)
 
     supervisor_agent = create_team_supervisor(
-        current_state.llm,
         SUPERVISOR_SYSTEM,
-        [VIDEOARCHIVE, ADOBEHELP],
+        [VIDEOARCHIVE, ADOBEHELP], config=config
     )
 
-    ai_graph = langgraph.graph.StateGraph(PsTutsTeamState)
+    ai_graph = StateGraph(PsTutsTeamState, config_schema=Configuration)
 
     ai_graph.add_node(VIDEOARCHIVE, rag_node)
     ai_graph.add_node(ADOBEHELP, adobe_help_node)
     ai_graph.add_node("supervisor", supervisor_agent)
 
-    edges = [
-        [VIDEOARCHIVE, "supervisor"],
-        [ADOBEHELP, "supervisor"],
-    ]
-
-    [ai_graph.add_edge(*p) for p in edges]
+    ai_graph.add_edge(VIDEOARCHIVE, "supervisor")
+    ai_graph.add_edge(ADOBEHELP, "supervisor")
 
     ai_graph.add_conditional_edges(
         "supervisor",
@@ -273,7 +244,7 @@ async def build_the_graph(current_state: ApplicationState):
         {
             VIDEOARCHIVE: VIDEOARCHIVE,
             ADOBEHELP: ADOBEHELP,
-            "FINISH": langgraph.graph.END,
+            "FINISH": END,
         },
     )
 
@@ -283,4 +254,5 @@ async def build_the_graph(current_state: ApplicationState):
 
 
 # Note: Cannot run build_the_graph() here as it requires current_state parameter
-# graph, _ = asyncio.run(build_the_graph())
+db = initialize_datastore(lambda _ : logging.info("Database initialized"))
+graph = asyncio.run( build_the_graph(db) )

pstuts_rag/pstuts_rag/state.py

@@ -1,24 +1,80 @@
-from dataclasses import dataclass, field
+from pydantic import BaseModel, Field
 from langchain_core.messages import BaseMessage
-
+from typing import List, Optional, Tuple, Dict, Annotated
 
 import operator
-from typing import Annotated, List, Tuple, TypedDict
 
 
-class PsTutsTeamState(TypedDict):
-    messages: Annotated[List[BaseMessage], operator.add]
-    team_members: List[str]
-    next: str
+class PsTutsTeamState(BaseModel):
+    """State management for PsTuts team workflow orchestration."""
+
+    messages: Annotated[List[BaseMessage], operator.add] = Field(
+        default_factory=list,
+        description="Accumulated list of messages exchanged during team collaboration",
+    )
+    team_members: List[str] = Field(
+        default_factory=list,
+        description="List of active team member identifiers participating in the workflow",
+    )
+    next: str = Field(
+        default="",
+        description="Identifier of the next team member or process step to execute",
+    )
+
+
+class StateInput(BaseModel):
+    """Input state for processing user queries and requests."""
+
+    input_query: str = Field(
+        description="The user's input query or topic to be processed by the system",
+    )
+
+
+class StateOutput(BaseModel):
+    """Output state containing processed results and references."""
+
+    output_content: str = Field(
+        default="",
+        description="The final generated content or report based on the input query",
+    )
+    video_references: List[Tuple[str, float]] = Field(
+        default_factory=list,
+        description="List of video references with starting timestamp",
+    )
+    url_references: List[str] = Field(
+        default_factory=list,
+        description="List of URL references related to the generated content",
+    )
+
+
+class RAGInput(BaseModel):
+    """Input for Retrieval-Augmented Generation processing."""
+
+    query: str = Field(
+        description="The search query to be processed by the RAG system",
+    )
+
+
+class VideoSegment(BaseModel):
+    """Summarized video segment"""
 
+    path: str = Field(default="", description="Path to video file")
+    summary: str = Field(default="", description="Summary of the transcript")
+    start: Optional[float] = Field(description="Start time of the segment")
+    stop: Optional[float] = Field(description="End time of the segment")
 
-@dataclass(kw_only=True)
-class StateInput:
-    input_query: str = field(default=None)  # Report topic
 
+class RAGOutput(BaseModel):
+    """Output from Retrieval-Augmented Generation processing."""
 
-@dataclass(kw_only=True)
-class StateOutput:
-    output_content: str = field(default=None)  # Final report
-    video_references: List[Tuple[str, float]] = field(default=None)
-    url_references: List[str] = field(default=None)
+    query: str = Field(
+        default="", description="The original query that was processed"
+    )
+    content: str = Field(
+        default="",
+        description="The generated content based on retrieved documents and query",
+    )
+    context: List[Dict[str, VideoSegment]] = Field(
+        default_factory=list,
+        description="Retrieved document context with relevance scores and metadata",
+    )

pyproject.toml

@@ -12,7 +12,7 @@ dependencies = [
     "langchain-experimental>=0.3.4",
     "langchain-openai",
     "langchain-qdrant>=0.2.0",
-    "langgraph>=0.4.3",
+    "langgraph>=0.2.55",
     "qdrant-client>=1.8.0",
     # API integration
     "chainlit==2.0.4",