Role purpose
The AI Agent Developer designs and builds the AI agents and assistants that make VisionONE easier to use, sell and support. This includes a voice/chat assistant for the website, in-product AI reporting and assistance, and automation across ERP, payroll and compliance workflows. You will turn business problems into reliable, safe, production-grade AI features — owning the full loop from prompt and retrieval design to deployment, evaluation and iteration.
Key responsibilities
AI agent & assistant development
- Design and build conversational and voice AI agents, including a website assistant, pre-sales and support bots, and WhatsApp-based automation.
- Implement natural, on-brand interactions with clear scope, fallbacks and human hand-off.
LLM integration & orchestration
- Integrate LLM APIs and build orchestration using retrieval-augmented generation (RAG), tool/function calling and structured outputs.
- Build and maintain knowledge bases and vector retrieval over VisionONE product, support and compliance content for India and Kenya, including GST/TDS/payroll topics and eTIMS/KRA workflows.
Automation & in-product AI
- Develop AI-assisted reporting, summarisation and workflow automation within the ERP environment, including finance, payroll and statutory workflows relevant to India and Kenya.
- Identify repetitive, high-volume tasks that AI can accelerate and build safe, measurable solutions.
Prompt engineering & evaluation
- Design prompts, guardrails and system instructions that keep outputs accurate, safe and on-brand.
- Build evaluation sets and test harnesses; measure quality, latency and cost, and iterate based on results.
Safety, guardrails & compliance
- Implement safeguards including clear AI disclosure, consent for recording/transcription, sensitive-data masking, authentication before accessing customer data and immediate human hand-off.
- Ensure AI features do not present legal, tax or compliance guarantees; use appropriate sourcing and human hand-off for statutory topics; and follow applicable data-protection requirements in India and Kenya.
Engineering, monitoring & collaboration
- Build supporting services and APIs and handle authentication, webhooks, rate limits and background processing.
- Log and monitor agent behaviour; set up alerting for failures, unsafe responses or drift and iterate from real usage.
- Work closely with product, web and support teams to embed AI where it adds measurable value.
Required qualifications & experience
- Degree in Computer Science, Data Science, AI/ML or a related field — or equivalent demonstrable experience.
- 2–4 years in software development, including at least 1 year building with LLMs / AI in real projects.
- Examples of AI features, agents or automations you have shipped, such as work samples, repositories or demos.
Technical skills
- Languages: strong Python; comfort with JavaScript/TypeScript.
- LLMs: hands-on integration of LLM APIs such as OpenAI, Anthropic, Groq or similar; agent/RAG pipelines using established frameworks or custom code.
- Retrieval: embeddings and vector stores such as pgvector, Chroma, FAISS or Pinecone; chunking and knowledge-base design.
- APIs & backend: REST, webhooks, authentication, async processing, queues and SQL databases.
- Ops: Docker, cloud deployment, CI/CD, secrets management, logging and monitoring.
- Practice: prompt engineering, evaluation methods and awareness of cost, latency and safety trade-offs.
Preferred / nice to have
- Voice AI / speech technologies such as STT/TTS and telephony or conversational platforms.
- WhatsApp Cloud API and messaging integrations.
- Familiarity with C#/.NET, SQL Server and enterprise/ERP data models.
- Angular exposure for embedding AI features in the product UI.
Key competencies
- Strong problem decomposition
- Pragmatism and shipping mindset
- Rigour around safety and evaluation
- Curiosity and fast learning
- Clear written communication
- Collaboration across product and business teams
What success looks like in the first 12 months
- First 30–60 days: understand the product and data; ship a first useful assistant or automation with appropriate guardrails.
- By 90 days: operate an in-production agent with measurable resolution/containment, monitoring and an evaluation loop in place.
- Ongoing: improve answer accuracy and user satisfaction, control latency and cost per interaction, grow a library of safe reusable AI components and maintain clean hand-off to humans when needed.