LLM Chatbot for HRMS-CRM
Budget: $250 – $750 USD
I want to embed a large-language-model chatbot directly inside both our in-house HRMS and CRM portals. All conversational context and answers must come from the data already stored on our own servers, then be returned to users in natural language.
Main goal
• When a staff member is inside the HRMS or a sales rep is inside the CRM, a side-panel chat should let them ask questions such as “summarise this employee’s last three performance reviews” or “give me a quarterly pipeline report in two bullet points”. The bot should instantly read the relevant tables, generate the summary or report, and show a reference to the source records so the response is auditable.
Scope of work
• Select or build an LLM solution (e.g., GPT-4 via API, open-source models like Llama 2, or a custom fine-tuned model) that can run securely against our data layer.
• Create middleware to map user prompts to our database schema, retrieve only the needed records, and feed them into the model.
• Expose the chat UI inside both web apps using the same design language we already use (React front-end).
• Implement role-based access so the bot never leaks data across departments.
• Return answers primarily in concise summaries and reports; leave room for future expansion to other generative tasks.
• Deliver annotated code, deployment scripts (Docker/K8s), and basic user documentation.
Acceptance criteria
1. Queries return accurate summaries within three seconds for typical record sizes.
2. No PII is exposed outside the user’s permission level in controlled tests.
3. Solution installs end-to-end in our staging environment with a single command and passes our integration test suite.
I will provide database ERDs, API keys (if we use a third-party model), and a staging server. Let me know which model architecture, vector store (if any), and security approach you would choose, plus a realistic timeline for delivery.
Main goal
• When a staff member is inside the HRMS or a sales rep is inside the CRM, a side-panel chat should let them ask questions such as “summarise this employee’s last three performance reviews” or “give me a quarterly pipeline report in two bullet points”. The bot should instantly read the relevant tables, generate the summary or report, and show a reference to the source records so the response is auditable.
Scope of work
• Select or build an LLM solution (e.g., GPT-4 via API, open-source models like Llama 2, or a custom fine-tuned model) that can run securely against our data layer.
• Create middleware to map user prompts to our database schema, retrieve only the needed records, and feed them into the model.
• Expose the chat UI inside both web apps using the same design language we already use (React front-end).
• Implement role-based access so the bot never leaks data across departments.
• Return answers primarily in concise summaries and reports; leave room for future expansion to other generative tasks.
• Deliver annotated code, deployment scripts (Docker/K8s), and basic user documentation.
Acceptance criteria
1. Queries return accurate summaries within three seconds for typical record sizes.
2. No PII is exposed outside the user’s permission level in controlled tests.
3. Solution installs end-to-end in our staging environment with a single command and passes our integration test suite.
I will provide database ERDs, API keys (if we use a third-party model), and a staging server. Let me know which model architecture, vector store (if any), and security approach you would choose, plus a realistic timeline for delivery.