Abhishek rajput profile photo 1 on Tal

Abhishek rajput

System Engineer | Backend, AI & Gen AI | From R&D to Deployment | RAG, Multi-Agent Pipelines, MCP Automation, Telemetry Systems

Ahmedabad

2+ years of work experience

CreatorOS
Jun 20261 moJul 2026

Ahmedabad

Designed, built, and deployed a full-stack UPI-native bookings and payments SaaS solo in 19 days, from empty repo to live. Engineered for correctness under money and concurrency: an append-only, event-sourced payment ledger (immutability enforced by Postgres triggers), a single-writer processor, and a partial unique index that makes the booking INSERT itself the availability check so double-booking is structurally impossible with no race condition. Idempotent webhooks, timeout-to-reconciliation handling, and a reconciliation sweep against the Razorpay API. 16,100 LOC across 13 tables, backed by 394 tests against real Postgres in CI. Stack: TypeScript, Next.js, Supabase (PostgreSQL).
Jun 20261 moJul 2026

Ahmedabad

Built and open-sourced a security/governance layer for AI agents on Anthropic's MCP. Instead of an agent holding standing credentials, Broker mints short-lived, policy-scoped tokens per tool call, with policy checked at mint time so a denied action never receives a credential in the first place (deny-at-mint). Signs GitHub App installation tokens (RS256) per request, maintains an append-only JSONL audit trail, and integrates directly with Claude Desktop as an MCP server. Documented honestly with a Limitations section. Stack: Python, MCP, GitHub App JWT, RS256.
Sep 20259 moJun 2026

Ahmedabad

Designed and built a production-grade RAG pipeline end to end. Query flow: contextualization and intent routing (predefined answers for greetings/summaries cut LLM calls ~30%), multi-query expansion, semantic retrieval over Qdrant, cross-encoder reranking, and context assembly with page-level citations answers streamed live over WebSocket. Self-hosted HuggingFace TEI embeddings (bge-large) for zero per-token embedding cost; tiered LLM inference via Groq. Engineered for failure: circuit breakers, backpressure, graceful degradation, and a metrics endpoint tracking p95/p99 latency and per-token cost. Stack: Python, Django, DRF, Celery, Redis, Channels, PostgreSQL, Qdrant, Next.js.
NEXUS AUTOMECH PRIVATE LIMITED
Apr 20241 yr 5 mosAug 2025

Ahmedabad, Gujarat, India

Co-led the 2-person software team for all robotics and IIoT projects within a 4-person cross-functional hardware/software team. Architected and delivered two major systems: an autonomous robotics platform (payload, ROS2, SLAM) and a scalable, multi-tenant IIoT backend (60k+ tags, 500+ users). Specialized in system integration, backend architecture (Python, Django, Celery, Redis), and real-time sensor fusion (YOLOv8 + UWB), achieving a 3x latency reduction on a core robotics feature. Drove projects from initial R&D and hardware selection through final production-ready deployment, demonstrating strong ownership and problem-solving skills.
Rapidops Inc.
Aug 20227 mosFeb 2023

Bhaskaracharya Institute For Space Applications and Geo-Informatics
Jun 20223 mosAug 2022

Gandhinagar, Gujarat, India

GitHub
@Abhishekvoid
LeetCode
@ab14609a

Abhishek’s AI Stack

antigravity
antigravity
claude code
claude code
codex
codex
gemini
gemini
langchain
langchain
langgraph
langgraph

Education

SILVER OAK UNIVERSITY

Bachelor of Technology - BTech, Computer Science

SILVER OAK UNIVERSITY

20213 yrs2024

Join Abhishek and 10k+ other professionals on

Tal Boss

Tal is the app where bosses meet engineers & builders directly, without middlemen

Download Tal Boss

Abhishek rajput on Tal

Headline

System Engineer | Backend, AI & Gen AI | From R&D to Deployment | RAG, Multi-Agent Pipelines, MCP Automation, Telemetry Systems

About

I build production AI systems end to end RAG pipelines, AI agents, and the backend infrastructure behind them. Recent work (all designed, built, and shipped solo): • RagBag production RAG pipeline: Qdrant retrieval, cross-encoder reranking, intent routing, streamed responses, observability. • Broker MCP open-source security layer for AI agents on the Model Context Protocol (deny-at-mint, scoped tokens, full audit). • CreatorOS bookings + payments platform taken from idea to deployed in 19 days. Before this, 16 months at Nexus Automech taking robotics and IIoT projects from R&D to production including a 3x latency reduction and a multi-tenant IIoT platform (60k+ tags, 500+ users) working directly with clients and hardware vendors. I thrive in high-ownership roles where I take a problem from R&D through production. Open to AI/backend engineering roles.

Location

Ahmedabad

AI stack

antigravity

claude code

codex

gemini

langchain

langgraph

Work experience

Creator / solo developer at CreatorOS, Jun 2026 - Jul 2026 · 1 mo, Ahmedabad, Designed, built, and deployed a full-stack UPI-native bookings and payments SaaS solo in 19 days, from empty repo to live. Engineered for correctness under money and concurrency: an append-only, event-sourced payment ledger (immutability enforced by Postgres triggers), a single-writer processor, and a partial unique index that makes the booking INSERT itself the availability check so double-booking is structurally impossible with no race condition. Idempotent webhooks, timeout-to-reconciliation handling, and a reconciliation sweep against the Razorpay API. 16,100 LOC across 13 tables, backed by 394 tests against real Postgres in CI. Stack: TypeScript, Next.js, Supabase (PostgreSQL).

Creator / solo developer at MCP broker, Jun 2026 - Jul 2026 · 1 mo, Ahmedabad, Built and open-sourced a security/governance layer for AI agents on Anthropic's MCP. Instead of an agent holding standing credentials, Broker mints short-lived, policy-scoped tokens per tool call, with policy checked at mint time so a denied action never receives a credential in the first place (deny-at-mint). Signs GitHub App installation tokens (RS256) per request, maintains an append-only JSONL audit trail, and integrates directly with Claude Desktop as an MCP server. Documented honestly with a Limitations section. Stack: Python, MCP, GitHub App JWT, RS256.

Creator / solo developer at Ragbag, Sep 2025 - Jun 2026 · 9 mo, Ahmedabad, Designed and built a production-grade RAG pipeline end to end. Query flow: contextualization and intent routing (predefined answers for greetings/summaries cut LLM calls ~30%), multi-query expansion, semantic retrieval over Qdrant, cross-encoder reranking, and context assembly with page-level citations answers streamed live over WebSocket. Self-hosted HuggingFace TEI embeddings (bge-large) for zero per-token embedding cost; tiered LLM inference via Groq. Engineered for failure: circuit breakers, backpressure, graceful degradation, and a metrics endpoint tracking p95/p99 latency and per-token cost. Stack: Python, Django, DRF, Celery, Redis, Channels, PostgreSQL, Qdrant, Next.js.

Software Engineer at NEXUS AUTOMECH PRIVATE LIMITED, Apr 2024 - Aug 2025 · 1 yr 5 mos, Ahmedabad, Gujarat, India, Co-led the 2-person software team for all robotics and IIoT projects within a 4-person cross-functional hardware/software team. Architected and delivered two major systems: an autonomous robotics platform (payload, ROS2, SLAM) and a scalable, multi-tenant IIoT backend (60k+ tags, 500+ users). Specialized in system integration, backend architecture (Python, Django, Celery, Redis), and real-time sensor fusion (YOLOv8 + UWB), achieving a 3x latency reduction on a core robotics feature. Drove projects from initial R&D and hardware selection through final production-ready deployment, demonstrating strong ownership and problem-solving skills.

Software Engineer at Rapidops Inc., Aug 2022 - Feb 2023 · 7 mos

AI Software Engineer at Bhaskaracharya Institute For Space Applications and Geo-Informatics, Jun 2022 - Aug 2022 · 3 mos, Gandhinagar, Gujarat, India

Education

Bachelor of Technology - BTech, Computer Science, SILVER OAK UNIVERSITY

Public links

github: https://github.com/Abhishekvoid

leetcode: https://leetcode.com/ab14609a