About Neural Network®
Where advisory recommendations get proven
Neural Network® is an independent, founder-led enterprise advisory firm headquartered in Pune with active engagements across India, the UAE, Saudi Arabia, Qatar, and Bahrain. We partner with GCCs (Global Capability Centres) and enterprise engineering organisations, including Fortune 100 companies in Banking, Healthcare, Technology, and Government, to close the measurable gap between AI investment and verified business outcomes.
Our five practice areas, Agentic AI Scaling, AI Governance & Compliance, Developer Productivity, Data Engineering & LLMOps, and AI FinOps & Token Economics, treat AI, Cloud, and Data as one integrated system. Every engagement opens with a baseline and closes with a verified delta. No vendor referral fees. No tool resale. Pure independent advisory.
When our advisory recommendations leave the whiteboard, this role is responsible for proving they work.
The Work, End To End
One pipeline, owned start to finish
Every engagement at Neural Network® ends with a verified delta, a measurable outcome the client can take to their board. This role sits at the technical edge of that process: taking an advisory concept and turning it into something demonstrable, testable, and architecturally sound enough to survive a CISO review. You will not hand off between stages. You own the full prototype.
- Client Problem
- AI Workflow Design
- API Integration
- LLM Selection & Prompting
- Data Pipeline
- Automation Layer
- Evaluation & Testing
- Working Prototype
What You Will Build & Do
Four disciplines, one owned pipeline
01 • Architecture
Workflow & prototyping
- Translate advisory engagement scopes into technical prototype architectures, given a client problem, define the AI workflow before writing a line of code
- Build working prototypes demonstrating Agentic AI, RAG pipelines, LLM-powered automation, and AI governance tooling on Azure (AI Agent Service, Azure OpenAI) and AWS (Bedrock Agents)
- Design multi-agent workflows with human-in-the-loop boundaries, nonhuman identity controls, and cost containment, not demos, systems that could survive a production readiness review
- Prototype AI FinOps instrumentation: token consumption tracking, semantic caching layers, and inference cost dashboards auditable by a finance team
02 • Integration
LLMs & prompt engineering
- Integrate LLM APIs (OpenAI, Azure OpenAI, Anthropic Claude, AWS Bedrock) using Python and FastAPI, as governed components with input validation, output evaluation, and cost tracking, not black-box calls
- Design and test prompt architectures for enterprise use cases: structured extraction, summarisation, compliance classification, code review, and RAG
- Evaluate LLM output quality systematically: build evaluation harnesses, define metrics, produce results defensible to a technical reviewer
- Research and benchmark model selection: GPT-4o versus Claude 3.5 versus a fine-tuned smaller model, with cost and latency implications quantified
03 • Data
RAG & pipeline engineering
- Build RAG pipelines: document ingestion, chunking strategies, vector database selection (Pinecone, Azure AI Search, pgvector), embedding model selection, retrieval evaluation
- Write clean, structured SQL and Python data pipelines, data quality, schema validation, and lineage tracking included, not bolted on afterward
- Implement semantic firewall and guardrail layers on LLM pipelines: input filtering, output validation, PII detection, compliance audit logging
- Instrument data pipelines to surface AI FinOps metrics: token spend by workflow, retrieval latency, cache hit rates, model utilisation
04 • Infrastructure
Cloud & deployment
- Deploy prototypes on Azure and AWS: containerised with Docker, version-controlled on GitHub, with environment configurations a second engineer could reproduce without a walkthrough
- Work with Azure OpenAI Service, Azure AI Agent Service, Azure Monitor, and Azure Key Vault at a level beyond documentation-reading
- Apply cloud cost governance from the start: right-sized compute, idle resource controls, spend visibility baked into the architecture
- Document deployment architecture at a level a client's cloud architect could review and challenge, client-grade documentation, not internal notes
05 • Evaluation, Testing & Governance Instrumentation
- Build evaluation and testing frameworks: unit tests for pipeline components, integration tests for end-to-end workflows, LLM output evaluation suites with defined pass/fail criteria
- Instrument prototypes with governance telemetry: audit logs, model version tracking, prompt versioning, and output lineage, the artefacts a compliance team needs before production
- Produce technical documentation supporting our advisory deliverables: architecture diagrams, data flow maps, cost models, evaluation reports
- Identify failure modes before they are discovered in review: adversarial prompting, retrieval failures, hallucination patterns, cost spikes, and document mitigations
What You Will Learn
Production-credible, not demo-ready
- How to architect and build AI systems that are production-credible, not just demo-ready, the difference between a working prototype and one that actually survives enterprise scrutiny
- The full AI engineering stack at an advisory level: LLM APIs, RAG architecture, vector databases, Agentic AI on Azure and AWS Bedrock, LLMOps instrumentation, and AI FinOps cost controls
- How enterprise AI governance is built into a system from architecture, not retrofitted, and what a CISO actually looks for in a technical review
- Cloud and AI as a single cost system: how token economics, compute costs, and data egress combine into a bill most finance teams have never properly audited, and how to model and control it
- How advisory recommendations translate into technical artefacts that GCC technology heads and CTOs can evaluate, challenge, and eventually deploy, the gap between slide and system
Who This Suits
Academic background and readiness
Any of the following:
B.Tech / B.E., Computer Science, IT, Electronics, or equivalent
MCA or M.Tech with a strong software engineering foundation
BSc Computer Science with demonstrated engineering depth beyond coursework
This Is Intermediate to Advanced. You Are Ready If You
- Have already built something non-trivial with an LLM API, not a tutorial chatbot, but a system with a defined input, a structured processing layer, and a measurable output
- Write Python that other engineers can read without a walkthrough: clean, typed, documented, with error handling that is not a bare except block
- Understand what RAG is, why naive chunking fails in enterprise document retrieval, and at least one approach to improving it, and can explain that without reaching for a blog post
- Can look at a cloud bill and identify where the spend is coming from at a service level, not just "it's expensive" but which specific workflow is driving the cost
- Have used Docker to containerise something you built and deployed it somewhere that isn't your laptop
- Have opinions about LLM evaluation, you have thought about what "the model gave a good answer" actually means in a production context, and tried to measure it
- Are comfortable working at the edge of what you know, researching, attempting, failing, and iterating, without requiring a fully specified brief before you start
This Role Will Not Suit You If
- Your primary experience is frontend development, mobile apps, or data science without an engineering foundation
- You have used LLMs through a UI (ChatGPT, Copilot) but not integrated them programmatically into a system you built
- You need a defined task list before you can produce output, the work here is often ambiguous by design
- You are not based in Pune
Technical Stack You Will Work With
The tools, by category
| Category | Technologies |
|---|---|
| Language | Python (primary), SQL |
| API Framework | FastAPI |
| LLM APIs | Azure OpenAI, OpenAI API, Anthropic Claude API, AWS Bedrock |
| RAG & Vector | Azure AI Search, Pinecone, pgvector, LangChain / LlamaIndex |
| Cloud | Azure (primary), AWS (secondary) |
| Containerisation | Docker |
| Version Control | GitHub |
| AI Agents | Azure AI Agent Service, AWS Bedrock Agents |
| Frontend | Basic, Streamlit or equivalent for prototype interfaces |
| Evaluation & Testing | Custom evaluation harnesses, pytest, LLM output benchmarking |
| Data | SQL pipelines, Delta Lake (conceptual), structured data ingestion |
Strongly Preferred
What sets an application apart
- At least one substantive project involving LLM API integration, RAG pipeline construction, or AI agent workflow, with code available for review (GitHub link required in application)
- Familiarity with Azure AI services beyond surface-level: Azure OpenAI Studio, Azure AI Agent Service, Azure Monitor, or Azure Key Vault
- Understanding of AI governance concepts: model registries, prompt versioning, audit logging, semantic firewalls, and compliance instrumentation
- Exposure to LLMOps tooling: MLflow, Weights & Biases, LangSmith, or equivalent evaluation and observability platforms
- Awareness of AI FinOps: token economics, inference cost modelling, semantic caching, and GPU right-sizing
- Knowledge of enterprise security fundamentals relevant to AI systems: nonhuman identity management, least-privilege API access, secrets management, and data residency constraints
What You Get
Real systems, not sandboxed tutorials
- Work directly with Rahoul Joshii, Founder & Principal Advisor, you will build systems that support real enterprise advisory engagements, not internal tooling no one uses
- Hands-on experience with Azure AI Agent Service, AWS Bedrock Agents, and the full RAG-to-production stack on live advisory problems, not sandboxed tutorials
- A portfolio of working AI prototypes built for real enterprise use cases in Banking, Healthcare, and GCC Technology sectors in India and the Gulf
- Competitive stipend
- Certificate of Completion and a detailed, specific performance reference (LinkedIn and written) from the Principal Advisor
- Meaningful consideration for a pre-placement offer or extended engagement based on the quality and production-readiness of work delivered
Location & Logistics
The practical details
- Office
- World Trade Centre, Unit No. 801, 8th Floor, Dholepatil Farms Road, Pune, Maharashtra, 411014
- Mode
- On-Site, Pune (no remote option)
- Duration
- Full-time role (Graduate / Fresher)
- Start Date
- Immediate
- Stipend
- Competitive