Codewits Solutions Pvt. Ltd.
Website:
codewits.in
Job details:
Experience: 10+ Years (including ~2 years relevant agentic AI experience)
Skills required:
Agent & sub‑agent architecture. Ability to design multi‑agent systems with planners, workers, and specialized sub‑agents, including clear hand‑off protocols and failure‑handling.
Around 2 years of relevant experience building and operating agentic workflows in real projects or products. MCP servers and integrations. Hands‑on experience designing and implementing MCP servers that expose internal APIs, databases, and workflows to agents in a secure and observable way.
Practical knowledge of MCP client integration, transports (e.g., STDIO/HTTP), authentication/authorization, and managing connections to multiple MCP servers. Context management & data access
Strong understanding of context engineering: chunking, summarization, windowing, and context isolation for long‑running or multi‑step agent flows. Ability to choose and justify between MCP‑exposed tools, private data stores, and RAG/vector search for internal data, considering latency, security, and governance.
Prompting, safety, and guardrails. Experience designing robust prompts for tools and agents, including patterns to mitigate prompt injection, prompt stuffing, and data‑exfiltration risks.
Familiarity with permissioned tools, human‑in‑the‑loop reviews, and policy checks before sensitive actions (e.g., code changes, production-impacting calls). Autonomous workflows & Ralph loops
Experience setting up Ralph‑style or similar long‑running coding loops that iteratively modify codebases, gated by tests, linters, and explicit stop conditions.
Skills in observing, logging, and debugging agent runs (traces, run logs, metrics) and iterating on prompts/skills/MCP design based on real‑world feedback.
Soft Skills:
Excellent communication, ability to explain trade‑offs to engineers and stakeholders, and a practical approach to balancing quality, safety, and delivery speed.
Click on Apply to know more.