Senior AI Engineer — FYNDNA AI Platform (FYNAI)

Experience: ~10 years hands-on software engineering, including 1-2 years building/shipping production LLM or ML systems Location: Mumbai or Pune

About the Role

FYNDNA is building a production AI layer (FYNAI) inside a core lending/banking platform: MCP-based code generation and governance tools, a natural-language-to-SQL pipeline over Databricks, AI-powered pricing/rules engines, and an in-house fine-tuned Small Language Model (SLM) replacing external LLM calls for cost, latency, and data-residency reasons. You’ll own the hard problems in that layer: model integration, fine-tuning, agentic tool orchestration, deployment, and making all of it reliable enough to sit in a regulated lending workflow.

What You’ll Work On

·      fynai-mcp — the platform’s MCP-based orchestration layer. One core capability is the NL-to-SQL/GraphQL pipeline: improve query routing, grounding, and safety for questions hitting Databricks SQL and GraphQL over live loan data

·      ai-loans-slm — fine-tune and evaluate a domain-specific SLM (LoRA/PEFT adapters) trained on schema + Q&A pairs, replacing Gemini/Claude calls with a locally-hosted model; own the training/eval loop and regression gates

·      MCP server fleet — extend and harden 7+ Model Context Protocol servers (code-assist, rules-creator, test-creator, code-governance, data-catalog, openapi-sync) spanning Python and Java/Spring AI

·      AI-powered pricing/rules services (fynai-epricing, fynai-pricing, fynai-rules) — rule/policy explanation and generation backed by pluggable Gemini/OpenAI/Claude

·      Work alongside the fynai-studio BMAD multi-agent design process, evaluating and improving how agentic workflows themselves get used for architecture, code, and test generation

·      Use tools like llama.cpp and vLLM to deploy SLMs

Required Skills & Experience

·      10 years hands-on software engineering; comfortable owning systems end-to-end, not just prototyping

·      1-2+ years building production LLM applications: RAG, prompt/context engineering, tool-calling/agentic patterns, and ideally direct experience with MCP (Model Context Protocol)

·      Hands-on fine-tuning experience: LoRA/PEFT, dataset curation, eval harness design, model regression testing. There’s a real SLM in production here, so you’ll need model management and MLOps experience too, e.g. vLLM

·      Strong Python (LLM tooling, FastAPI-style services) and enough Java/Spring Boot fluency to integrate AI into an existing Spring Boot/Gradle microservices estate (Spring AI experience a strong plus)

·      Experience with at least one major LLM provider API (Gemini, OpenAI, or Claude) in a pluggable/multi-provider architecture

·      Comfortable with the surrounding infra: PostgreSQL/YugabyteDB, Dapr, GraphQL, and ideally Databricks SQL

·      Can reason about safety/grounding for AI systems that touch financial data; this sits inside a regulated lending platform

Nice to Have

·      Prior BFSI / lending / core banking domain exposure

·      Experience with Temporal or similar durable workflow engines

·      Familiarity with multi-agent orchestration frameworks (BMAD or similar)

·      Experience shipping locally-hosted/on-prem models as a cost or compliance play, versus pure API consumption

Seniority Framing

At 10 years, this role expects fine-tuning experience already in hand, not learned on the job: opinions on eval design, a working sense of when fine-tuning beats RAG beats prompting, and the standing to push back on architecture rather than just execute it. Ownership of the SLM’s model lifecycle (train → eval → deploy → monitor drift) is the anchor deliverable.