Shard-Based RAG POC - 5M Token Training Set on Commodity Hardware
Summary
Built a working RAG proof of concept trained on ~5M tokens, using BERT for embeddings and Hugging Face datasets as the source corpus.
About
Highly accomplished AI Product Architect and Senior Software Engineer with 7+ years of experience, including 5 years in enterprise .NET microservices for international fintech clients and 1 year designing/shipping agentic AI workflows. Recognized as a two-time Outstanding Performer, I bridge stable enterprise stacks with modern AI orchestration (multi-model agents, MCP, fine-tuning, embeddings). Seeking AI Product Architect/Solutions Architect roles to leverage deep stack knowledge, business judgment, and hands-on AI engineering to deliver innovative, trustworthy AI solutions.
Work
Teamarcs Technologies Pvt. Ltd.
|Senior Software Engineer
Faridabad, India, India
→
Summary
Led a 5-person engineering team in developing microservices-based platforms for fintech clients, driving architectural transitions and integrating modern frontends.
Highlights
Recognized as a two-time Outstanding Performer (2023, 2024), the sole individual in the function to achieve this consecutive award for exceptional technical leadership and consistent delivery.
Led a 5-person engineering team in the full lifecycle delivery of microservices-based platforms (FRUIT WALA, WIZZ TUTOR), encompassing architecture, sprint planning, code review, and direct client technical engagement.
Architected and executed a 20-microservice project transition to a 4-modular monolith architecture within a one-month deadline, significantly streamlining system complexity and improving maintainability.
Integrated modern frontends (React, Blazor) with .NET Core backends, resolving complex challenges such as cross-origin authentication, real-time data freshness, and graceful degradation for robust user experiences.
Mentored junior developers on CI/CD fundamentals, code review best practices, and writing readable code, accelerating team ramp-up and fostering a high-performance culture.
Manufacturing & Tech Services (Self-Employed)
|Founder / Owner
India, India
→
Summary
Successfully managed a manufacturing unit with full P&L responsibility, overseeing vendors, finance, operations, and personnel to drive business growth.
Highlights
Managed a manufacturing unit with full P&L responsibility, overseeing vendors, finance, operations, and personnel, demonstrating credible product and architecture conversations with business stakeholders.
Translated commercial constraints into weekly operational decisions, showcasing a strong understanding of feature value vs. inference cost for AI Product Architect roles.
Techahead
|Software Engineer
→
Summary
Developed web applications and REST APIs using ASP.NET MVC, AngularJS, and Node.js, translating client requirements into shippable features.
Highlights
Developed and deployed web applications and REST APIs using ASP.NET MVC, AngularJS, and Node.js, directly translating client requirements into shippable, high-quality features.
Optimized performance across multiple live projects, notably reducing query times on slow MSSQL workloads through advanced indexing and query restructuring, enhancing user experience and system efficiency.
Successfully delivered 10 Live Projects from ideation to deployment across Social Apps, E-commerce, and Fintech industries, demonstrating end-to-end project ownership and impact.
Education
Continuing Education
→
AI Project Management (AIPM)
AI product strategy, agentic system design, and managing AI-led delivery.
CDAC, Noida IPU
Noida, Uttar Pradesh, India
→
Master of Computer Applications (MCA)
Computer Applications
Grade: 74%
DAVIM MDU
→
Bachelor of Computer Applications (BCA)
Computer Applications
Grade: 67%
Awards
Outstanding Performer
Awarded By
Teamarcs Technologies Pvt. Ltd.
Recognized for exceptional technical leadership and consistent delivery.
Outstanding Performer
Awarded By
Teamarcs Technologies Pvt. Ltd.
Recognized for exceptional technical leadership and consistent delivery.
Outstanding Performer
Awarded By
Techahead
Recognized for outstanding contributions and performance.
Languages
English
Hindi
Skills
Agentic AI Architecture
Multi-model orchestration, MCP tool routing, Agent workflows, Memory & context management, Intent classification, Deterministic-vs-generative routing.
LLM Engineering
Local LLM serving (Llama 3.1 8B GGUF, xLAM 1B), GGUF quantisation trade-offs, Fine-tuning (domain data), Prompt engineering, RAG architectures, Evaluation.
Embeddings & Retrieval
BERT (SBert & Modern Bert with different Token Sizes) embeddings, Shard-based vector storage (.npy), Hugging Face datasets, JSONL reasoning datasets, Retrieval pipelines, Graph-based RAG design.
Backend Engineering
C#, ASP.NET Core, Web API, MVC, WCF, Microservices, LINQ, Entity Framework, Python, FastAPI.
AI Tooling & Integration
MCP protocol (Blender MCP, custom MCP servers), Groq API, pdfplumber, Markdown converters, Chart generation, PDF generation, SQL-grounded agents.
Data & Messaging
MSSQL (expert), PG SQL, PG Vector, Entity Framework.
Frontend & Delivery
Reactjs, Blazor, JavaScript, jQuery, AngularJS, ASP.NET Web Forms, Bootstrap.
Cloud & DevOps
AWS, CI/CD fundamentals, Jira, TFS.
Architecture Judgement
Docker, Kubernetes scheduling, Sync vs. Async I/O, OS-level cost of file ops, Algorithm selection.
Leadership & Delivery
Team leadership (5-person dev team), Mentoring, CI/CD adoption, Client requirements gathering, P&L ownership, Strategic planning, Stakeholder management, Process optimization, KPI tracking.
Projects
Financial Agentic Workflow - Multi-Model Orchestration with Non-Hallucinating Math
→
Summary
Designed and shipped an end-to-end agentic financial reasoning system via a FastAPI backend, integrating chat input and a client SQL database to answer questions grounded in live business data.
Chat-to-3D Learning Tool - Blender MCP + Groq
→
Summary
Built a learning tool generating Blender 3D scenes from chat messages or hand-drawn sketches using Blender MCP server and Groq API.