An AI HR Agent That Gave the People & Culture Team Their Time Back

About Cleo AI

Industry

AI HR Automation

Timeline

Less than 1 week

Technologies

Python, Customized Agentic Framework ( Based on LangChain), RAG Architecture, Multi-Provider LLM, Model Agnostic Inference, MCP Integration

At a 15–20 person company, HR should be simple. But "simple" doesn't mean frictionless. TechCare's People & Culture team was spending a significant chunk of every workweek answering the same five questions on Slack: What's my leave balance? Has my request been approved? Did my check-in go through? The data existed. The problem was access. Cleo is TechCare's in-house AI HR agent built to solve exactly that — an always-on, RAG-powered assistant that handles routine HR operations end to end, so the team can focus on people instead of process.

Performance Highlights

10+

Hours Saved Per Week

<2hr

Leave Approval Time (was ~1 full day)

0

Leave Balance Discrepancies

<1wk

Idea to Working Prototype

Solutions & Impact

At a 15–20 person company, HR should be simple. But "simple" doesn't mean frictionless. At TechCare, the People & Culture team was spending a significant chunk of every workweek answering the same five questions on Slack: What's my leave balance? Has my request been approved? Did my check-in go through?

The answers were always in the system. The problem was access, or rather the lack of it. Employees had no way to check their own data without messaging HR directly. Attendance was tracked manually, and leave approvals occasionally fell through the cracks when a manager had competing priorities.

The result was an estimated 8–12 hours of HR admin per week that required no human judgment, just time.

"Most Slack messages to HR were the same five questions, and the answers were always in the system. The data existed. The problem was access."

— TechCare P&C Team
Thumb

Why Off-the-Shelf Tools Weren't the Answer

TechCare had tried existing HR platforms, including Zoho. The integrations were limited, the workflows didn't match how a lean hybrid-remote team actually operated, and the team kept falling back on spreadsheets and direct messages to fill the gaps. Off-the-shelf solutions also came with enterprise-scale complexity that a team of 15–20 people simply didn't need.

The decision to build in-house was not about cost alone. It was about control — over the logic, the tone, the data, and the ability to evolve the tool as the company grows.

Meet Cleo

Cleo is TechCare's in-house AI HR agent, deployed on Slack and configurable for Telegram or any other preferred messaging stack. It was built by the TechCare team using a custom pipeline on top of FastAPI and LangChain, with a model-agnostic inference engine that runs across multiple LLM providers with a built-in fallback system.

At its core, Cleo uses a Retrieval-Augmented Generation (RAG) architecture, meaning every answer it gives is grounded in TechCare's own HR data, policy documents, and real-time database connections — not hallucinated from general knowledge. When Cleo cannot answer with sufficient confidence, a threshold mechanism triggers human-in-the-loop (HITL) escalation, automatically pulling in the HR admin.

From idea to beta-tested working prototype: less than one week.

Thumb

What Cleo Actually Does

Leave Management

Cleo handles leave requests end to end. In guided mode, it captures the request from the employee, notifies the reporting manager instantly, and updates the employee as soon as a decision is made.

In autopilot mode, it goes a step further. Cleo applies the HR rule book on its own, checking leave availability and policy compliance before approving or declining the request.

What changed most is the follow-through. If a manager does not respond within two business days, Cleo escalates automatically. Before Cleo, there was no system to catch that delay, and requests often stayed unresolved until someone manually followed up.

Attendance Tracking

Daily attendance, which was previously logged manually, now syncs directly with the HR dashboard in real time. There is no manual entry step, and therefore no risk of miscounts — especially around half-days, which had been a recurring source of spreadsheet errors.

Employee Self-Service

Employees can ask Cleo anything HR related at any time, including leave balances, policy questions, check-in confirmations, and onboarding guidance. New hires no longer need to ping HR for basic orientation questions in their first week. Cleo handles these queries and flags anything that genuinely requires a human.

Thumb

Admin Dashboard

HR administrators configure Cleo's behaviour, rules, and escalation logic through a dedicated dashboard. Full employee profiles can be uploaded into the system, and Cleo generates context-aware responses based on the HR rule book from day one. Onboarding introductions and shoutouts can also be pushed directly to Slack or any connected channel from the same dashboard.

The Technical Architecture

Layer
Approach
Framework
FastAPI + LangChain-based custom pipeline
LLM Inference
Model-agnostic; multi-provider with fallback
Knowledge Retrieval
RAG over company documents, HR policy, and live database connections
Hallucination Prevention
Query confidence thresholds + HITL escalation
Integrations
API and MCP connections for dynamic, real-time data fetching
Data Privacy
Employee conversations stored only for contextual continuity; no cross-employee data leakage

The internal processing logic — how Cleo understands, routes, and responds to queries — is proprietary IP and not disclosed publicly. What TechCare can confirm is that the system includes fine-tunable skills, an evolving feedback loop for continuous reasoning improvement, and security guardrails built into the architecture from the ground up.

Results

Response Time

Routine queries such as leave balance checks, attendance status, and check-in confirmations are now resolved instantly. Approvals that previously took around one full working day now complete in under two hours, driven by automatic manager escalation.

Accuracy

Leave balance discrepancies — previously a recurring issue with manual spreadsheet tracking — have been eliminated. Real-time sync ensures that the data Cleo surfaces is always current.

Time Saved

The P&C team has reclaimed an estimated 10+ hours per week previously spent on low-value admin work. That time is now redirected toward hiring, culture initiatives, and people development — the work that actually moves the needle.

Onboarding Experience

New hires report a more structured and less confusing first week. Cleo answers the orientation questions that previously required chasing down a member of the HR team.

Scalability

At 15–20 people, 10 hours saved per week is meaningful. At 50, 100, or 200 employees, the same manual HR processes would either require proportionally more headcount or begin to break. Cleo scales without adding overhead.

Why Founders Should Pay Attention

As companies scale, operational complexity grows faster than headcount. Hiring more HR personnel solves part of the problem. Operational automation solves the root problem.

What Didn't Exist Before Cleo

Before Cleo

Icon
Employees had no self-serve access to their own HR data
Icon
Leave approval escalation had no automated mechanism
Icon
Attendance required manual daily entry
Icon
HR policy answers depended entirely on a human being available to respond
Icon
New joiners had no structured first-week support

After Cleo

Icon
Employees self-serve leave balances, policies, and check-ins any time
Icon
Manager non-response auto-escalates after two business days
Icon
Attendance syncs in real time — no manual entry, no miscounts
Icon
Policy queries answered instantly, grounded in actual HR documents
Icon
New hires get structured onboarding support from day one

Key Takeaways for Teams Considering a Similar Build

1

The Data Already Exists — The Problem Is Access

If your HR queries are repetitive, the answers are likely already in your system. An AI agent is primarily an access layer, not a knowledge creator.

2

RAG Is Non-Negotiable in an HR Context

Hallucinations in HR — such as wrong leave balances or incorrect policy interpretations — erode trust quickly. Grounding every response in your actual data is the baseline requirement.

3

Human-in-the-Loop Is Not a Fallback — It Is a Feature

Cleo is not designed to replace HR. It is designed to handle what does not need a human, so humans can focus on what does. The escalation threshold is a deliberate design choice, not a limitation.

4

Build for Your Workflow, Not the Tool's Workflow

Off-the-shelf platforms often require you to adapt your processes to their logic. An in-house build allows the system to adapt to how your team actually works.

5

The ROI Compounds With Headcount

At a small scale, HR automation is a quality-of-life improvement. At scale, it becomes a structural necessity.

What's Next

TechCare plans to include a Cleo-specific question in the next employee engagement survey to formalise satisfaction data. Informal feedback to date has been consistently positive, particularly around the elimination of the back-and-forth employees previously needed just to check routine information.

Cleo was never designed to replace HR teams. It was designed to remove the repetitive operational burden that prevents HR professionals from focusing on people. The result is not just faster workflows — it is a better employee experience, a more scalable operation, and a glimpse into how internal AI systems will quietly reshape the future of work.

FAQ

How long does it take to set up? 

The TechCare team went from idea to working prototype in under a week. For a new organisation, setup time depends on how structured your existing HR data and policy documentation already is. The cleaner your inputs, the faster Cleo gets up to speed.

What happens when Cleo doesn't know the answer? 

Cleo doesn't guess. When Cleo's confidence drops below a defined threshold, it automatically escalates to a human HR administrator. This is a deliberate design choice, the system is built to know its own limits.

Is our employee data safe? 

Conversations are stored only to maintain contextual continuity within a session. No employee can use Cleo to access another employee's data. Security guardrails are built into the architecture, not added on top of it.

Can other companies use Cleo?

Cleo was built in-house at TechCare. If your team is dealing with the same problem, HR time disappearing into repetitive admin, then get in touch. We're happy to have that conversation.

Other Case Studies

Works Thumb
Mental Health Care

Scaling Access & Value

Our healthcare platform project raised $550K, sustained success for 5+ years, and expanded reach with powerful mobile and web experiences.

Works Thumb
Education

Transforming Learning Platforms

Our dedicated tech teams delivered innovative tech solutions, optimizing operations and driving impressive cost savings and growth.

Works Thumb
Home Maintenance

Turning Yardly’s Website Into a Scalable Growth Engine

Yardly is a Canada-based yard care service provider offering year-round solutions including lawn maintenance, seasonal clean-ups, and snow removal across Edmonton and Calgary.

Works Thumb
E-commerce

E-commerce with a Mobile-First Experience

One of Bangladesh’s leading online bookstores and e-commerce platforms, serving millions of readers and shoppers across the country.

Works Thumb
Artificial Intelligence / Healthcare

AI-Driven Doctor Recommendation System

Created an AI-powered matching engine using vector embeddings to connect patients with the most relevant healthcare specialists in seconds.

Works Thumb
Artificial Intelligence / Productivity

AI K-Base Assistant for Multi-Stakeholder Teams

Built a Claude-powered chatbot that delivers real-time, personalized answers across teams, automating knowledge sharing and documentation.