Development of a high-load system with fine-tuning of LLM models

Budget: 7600 USD Deadline: 45 days

Look, there's a nuance here - this is not just a chatbot task, but a product search system where the cost of architectural mistakes is high =/

For the timeline, I would allocate the first safe phase for 45 days and 7600 USD - design, testing multimodal search on your products, a prototype agent in messengers, data schema, photo processing queue, search quality metrics, and a plan for model retraining. Full development will depend on the load and catalog, most likely this will be a separate budget starting from 20000 USD.

The vision for implementation is as follows -
> normalize the catalog, attributes, photos, and texts of products
> build multimodal search - vector index, ranking, filters, relevance checking
> create an agent in messengers that clarifies the request, shows options, and sends events to the admin panel
> separately assess search quality and data for retraining, otherwise the model will respond beautifully but sell weaker.

From you at the start, I need the product catalog, examples of photos, descriptions of target messengers, and at least a rough estimate of peak loads. Overall, it’s fine to start not with a large construction project, but with a verifiable phase - first prove the quality of search on real products.

I will clarify 2 things -
> what is the volume of the catalog - products, photos, languages, updates per day
> is retraining the LLM really necessary or can we start with multimodal embeddings, ranking, and RAG, leaving fine-tuning for after accumulating training examples?

From recent experience -
> https://business.ingello.com/vorfahr - AI service, working with photos, content generation, and product logic
> https://business.ingello.com/fractal - agent architecture and automation of complex workflows
> https://business.ingello.com/prime-eva - product data, manufacturing, integrations, and accounting

Main page of the team for FLH -
> https://systems-fl.ingello.com Expand

Budget: 1900 USD Deadline: 33 days

Good day. The task is clear: multimodal product search, where the request comes in both photo and text simultaneously, plus an assistant agent that lives in messengers and communicates with the user, and all of this should handle the load.

In terms of architecture, I see it like this. We build the multimodal part on shared embeddings for images and text, so that the photo and text request fall into the same vector space and ranking is done based on proximity. This is supported by a vector database (I have worked with Qdrant in production) and a re-ranking layer on top of the candidates. Fine-tuning is needed where the ready model misses your nomenclature: we continue training on your pairs of product and request, without touching everything indiscriminately.

I have close experience with agents in messengers: I created a voice AI assistant for clinics, where the LLM was the orchestrator of the dialogue with tool use and knowledge base search through RAG, and a platform with semantic search for content in Telegram. I handle the high-load part with asynchronous Python (FastAPI), queues, and caching of hot requests.

To aim for load and cost: what is the order of the catalog and how many requests at peak do you expect, and in which specific messengers should the agent reside? This affects the choice of model and infrastructure.

Budget: 12500 USD Deadline: 45 days

Hello, Nik. Are you ready to do the automation and the agents that will control/improve it? Message me privately, don't waste time.

Budget: 100 USD Deadline: 5 days

Hello.

I see that you need not just LLM integration, but a high-load multimodal search system with model fine-tuning, capable of processing photo and text queries through an AI assistant in messengers without loss of quality and speed.

I have worked on similar AI projects: I created AI consultants, automation systems based on ChatGPT/Claude, integrations through Make.com, Voiceflow, Chatfuel, and CRM ecosystems, where the key task was to handle large volumes of requests, data routing, and building a scalable architecture.

For such a project, it is critical to think through not only the fine-tuning of models but also the entire pipeline: image processing, vector search, RAG architecture, caching, task queues, load monitoring, and agent operation in messengers. One effective approach is to separate the search, inference, and communication layers into independent services for horizontal scaling.

Could you let me know if you plan to use ready-made open-source models (LLaMA, Qwen, Gemma) or are you considering training your own model for the product catalog?

Budget: 2500 USD Deadline: 30 days

Hello.

Strong AI systems start not with fine-tuning, but with the right architecture. I am ready to offer a solution that will work under load today and won't require rewriting tomorrow.

I would be happy to discuss the project.

Budget: 25 USD Deadline: 1 day

Hi Nick,

This is squarely my area: AI search, vision, and agent assistants in messengers. One honest reframe up front, since it changes the budget.

"Multimodal photo + text search" usually doesn't need LLM fine-tuning. The search core is image+text embeddings (CLIP-style) indexed in a vector DB (pgvector or
Qdrant) with hybrid filtering and a reranking pass. That gets strong relevance on your catalog without the cost and fragility of training. Real fine-tuning only
earns its keep if we measure generic embeddings underperforming on your specific products, and then it's targeted, not the whole core.

The LLM's actual job is the agent layer in the messenger: clarify the query, call the search tool, present product cards. That's function-calling/RAG, not the
retrieval engine.

How I'd build it:
- Ingest catalog (photos, text, attributes), embed, index in a vector DB
- Hybrid search (vector + attribute filters) with a reranker
- Agent in Telegram/WhatsApp via Bot API: photo or text in, ranked cards out
- Event-driven queue for photo processing and traffic spikes, so nothing drops under load

To de-risk it, I'd start with a short paid discovery + a prototype on a slice of your real catalog, measured with recall@k, before committing to the high-load build.
That proves quality on your data first.

Relevant work: I built a dashboard over a private GPU inference platform (FastAPI, vector memory, live model state), shipped production LLM features and vision-API
search, and event-driven messenger bots. Examples in private. 36 projects on Freelancehunt, all 100%, 10/10.

Three questions to scope it: catalog size, target load (requests/min), and which messenger first? And is there an existing catalog API? Expand

Budget: 165 USD Deadline: 1 day

Good day, Nick.

In brief:

Your service will receive a multimodal search capable of simultaneously processing photo and text queries through a personal agent in Telegram or WhatsApp. The search core based on a fine-tuned LLM and vision models will ensure high accuracy of results. We will enhance text search with a vector index pgvector to instantly find similar products even with incomplete descriptions. The system will operate on Kubernetes with automatic scaling, so peak loads will not affect speed. Integration through messengers will allow users to send photos and text in chat, and the assistant will return a selection of products from your catalog. Fine-tuning on your data guarantees accuracy unattainable for general models.

More details:

Multimodal search requires deep fine-tuning of the LLM for your catalog; otherwise, relevance will remain at the level of generic models. The search core will combine a base model with additional training on your data and a vector index pgvector for semantic search. The vision component will extract features from photos, while text will clarify the context—both streams will merge into a single ranked result. The architecture on Kubernetes with independent scaling of nodes (vision, text, ranking) will ensure stable latency. An event-driven architecture with message queues will protect against query loss during traffic spikes, and the user will receive product cards with direct links.

The next step is to increase search accuracy through a hybrid RAG approach with fine-tuning. This will allow the agent to instantly update knowledge about stock and prices without retraining the model, which is critical for a dynamic e-commerce catalog.

Later on, conversion will increase thanks to an analytical module that will track the most popular visual queries. You will gain a clear understanding of what users are searching for to optimize purchasing and marketing.

I am ready to discuss the details of the catalog, query volumes, and the choice of the base model for fine-tuning. Expand

Budget: 1300 USD Deadline: 45 days

We have experience in building high-load architectures and fine-tuning LLM for multimodal tasks. We implement the system through microservices in Python using vector databases for fast search and integrating messenger APIs for agent operation. We will ensure scalability and accuracy in processing requests for photos and text. We are ready to start designing the architecture.

Budget: 25000 USD Deadline: 35 days

We already have a practically ready similar solution for AI search and assistants in messengers, which can be quickly adapted and launched for your catalog ))
For the task, I would estimate the first working stage at 250,000 UAH and about 35 working days for a prototype with photo and text search, ranking, an agent, and basic integration into messengers.
Look, there’s a nuance here - for the industrial version, it’s crucial to check in advance !!the load, search quality, and catalog structure!!, otherwise fine-tuning can become an expensive toy without noticeable gains.
For implementation, I would go through a data layer, vector and text search, a separate ranking layer, agent scenarios, logging requests, and quality metrics on the service side.
From you, I need the catalog export, photos, descriptions, output rules, a list of messengers, and access to the test API.
I’d like to clarify two points - what is the size of the catalog and what is the target load in requests per minute?
Which messengers are needed for the first launch and is there already an API for the catalog?
Similar examples for AI and agents - https://business.ingello.com/vorfahr and https://business.ingello.com/fractal
As a close example for e-commerce and catalog logic - https://business.ingello.com/prime-eva
More about us for the exchange - https://systems-fl.ingello.com
In touch, we can discuss the details within the project =) Expand