# Tanmay Bohra **Full Stack Engineer · Consultant** — Mumbai, India Also written: Tanmay Sunil Bohra. Site: https://tanmaybohra.com I build systems in Go and TypeScript, from high-concurrency APIs to agentic AI pipelines. I find the manual work inside an organisation and ship the tool that removes it. Currently at Grant Thornton Bharat. **Skills:** Go, Java, TypeScript, NestJS, PostgreSQL, Agentic AI (LangGraph, n8n), Kafka, React, AWS, Docker, Next.js, Python, Tailwind, Astro, Flutter **Certifications:** AWS Certified Cloud Practitioner; Six Sigma Green Belt ## Professional projects - **Purpleboard** (Grant Thornton): internal AI-agent platform used across teams. NestJS and PostgreSQL on AWS, shipped through CI/CD. https://purpleboard.gtbharat.in - **Arova**: agentic accounting for Indian fund managers. Bills and bank statements go in, TDS and GST come out with a citable reason, and journals post straight into the client's Tally or ERPNext. Python, React, PostgreSQL, Tally. https://accounting.tanmaybohra.com - **AuditFlow**: define the audit checks, upload the documents, and an AI agent runs the workflow and returns a compliance report in minutes. Python, FastAPI, React, PostgreSQL. https://auditflow.tanmaybohra.com ## Agentic builds (proof of concept, not live) - **Procurement Agent, Zinc and Steel**: watches commodity prices for a manufacturer's procurement team and recommends how much to buy, when, from which country and what to push for in vendor negotiation, with probabilities instead of a single guess. - **Travel Refund Assurance**: audits airline cancellations an enterprise has already made, compares the refund received with the refund owed under fare rules and DGCA limits, and reconciles the GST and ITC effect. ## Work with me One hour, one video call, INR 599. You leave knowing exactly what to build, fix or do next. For students and early developers (career direction, architecture, resume review), people building a product or startup (stack choices, backend design, AI agent integration, cutting scope), and anyone stuck on a specific problem (Go, TypeScript, NestJS, system design, agentic pipelines). Booked and paid by a person on https://tanmaybohra.com via Razorpay. ## Personal projects - [Space Local](https://blogs.tanmaybohra.com/games/): pixel-art endless runner on a Mumbai local train in space, with a world leaderboard - [Axiom Research OS](https://axiom-research-os--tanmaybohra.replit.app/): research intelligence dashboard for M&A, built in the Replit Buildathon - [Database Internals](https://databaseinternals.tanmaybohra.com): storage engines, indexing and query execution - [Blog Bahadur](https://blogs.tanmaybohra.com): writing on backend systems, Go and agentic AI - [Binary Trees](https://binarytrees.tanmaybohra.com): interactive tree-traversal visualiser - [Formula 1](https://formula1.tanmaybohra.com): F1 race and driver data explorer - [Supercode](https://supercode.tanmaybohra.com): developer tooling experiments - SalonFlow: phone-first field sales CRM for salon distribution (client project, private) - Chullan Cosmetics: storefront for a professional makeup brand (client project, private) ## Previously RP Tech India (Flutter development and cybersecurity intern), Commudle (community, events and developer relations), Grootin (technical staff and content writing intern), Fllink (development, graphic design, relations and event management). ## Contact - Email: hello@tanmaybohra.com - GitHub: https://github.com/sabaka3560 - LinkedIn: https://www.linkedin.com/in/tanmay-bohra/ - Blog: https://blogs.tanmaybohra.com ## Latest writing - [Stop Picking the Smartest OCR Model. Pick the Right One.](https://blogs.tanmaybohra.com/posts/agentic-ocr-verification-finance-health/): Claude, Gemini, GPT-4o, Textract, Document AI, or self-hosted OCR: a real per-page cost and use-case breakdown for finance and health docume... - [Token Economics for Finance: What AI Automation Actually Costs in India (2026)](https://blogs.tanmaybohra.com/posts/ai-token-economics-finance-india-2026/): Exact rupee costs for AI-powered reimbursement, invoice processing, due diligence, and GST filing — no discounts, no grants, June 2026 price... - [Sakana Means Fish. Their New Model Is a School of Them.](https://blogs.tanmaybohra.com/posts/sakana-fugu-multi-agent-orchestration/): Sakana AI built a 7B conductor that orchestrates GPT-5.5, Claude, and Gemini. On GPQA-Diamond it beats Mythos 5. Here is exactly how it work... - [Most Consultants Are Eloquent. Few Are Clear.](https://blogs.tanmaybohra.com/posts/epic-framework-consultant-communication/): EPIC is the communication framework I use in every client session. Four moves: Express WHY, Point first, Illustrate with stories, Close on c... - [CrewAI Thinks in People. LangGraph Thinks in Process.](https://blogs.tanmaybohra.com/posts/crewai-vs-langgraph-mental-model/): CrewAI vs LangGraph isn't a capability debate — it's a mental model debate. CrewAI thinks in people. LangGraph in process. Here's which one... - [I Built a WebSocket Server Just to Show Three Progress Messages](https://blogs.tanmaybohra.com/posts/langgraph-invoke-to-production-story/): Three wrong turns building a LangGraph agent — chained functions, while loops, and WebSockets for progress — and the production fixes that a... - ['Respond Only With Valid JSON' Fails 1 in 10 Times](https://blogs.tanmaybohra.com/posts/stop-prompting-for-json-constrained-decoding/): Naive JSON prompting fails 5-15% in production. Every major LLM provider ships constrained decoding now — the model cannot violate your sche... - [A Jailbroken Claude Was Used to Attack Claude](https://blogs.tanmaybohra.com/posts/multi-agent-jailbreak-attack-surface/): Claude Fable 5 survived 1,000 hours of bug bounty testing. A multi-agent pack hunt bypassed it in 48 hours. Here's why classifiers can't see...