Step 01 of 04
Understand the problem first.
Before I open an editor, I talk to the people who will use the thing. What do they need, and what happens if it's wrong?
AI Engineer · Claude Certified Architect
The mystery of human existence lies not in just staying alive, but in finding something to live for.
Simple problems. Complex methods. Simple explanations.
I'm an AI engineer at Tata Consultancy Services. I build AI agents and the plumbing around them, so big companies can actually rely on them.
Credentials
Things you can check.
Certifications, research, and recognition. Each card opens the proof.
Claude Certified Architect, ProfessionalAnthropic · 2026Verify
Claude Certified Architect, FoundationsAnthropic · 2026Verify- Published on arXiv, indexed in NASA ADSACCIDENT @ CVPR 2026Verify
- HarvardX CSAI certificateHarvard Extension School · 2024See details
- Guinness World Record participantOnline programming lesson · 2021See details
- 8 awards at TCSTata Consultancy ServicesSee details
My process
How I like to work.
Step 02 of 04
Let code do the maths.
Models are great at reading and explaining. Anything that has to be exact, like money, rules, or IDs, goes through tested code instead.
Step 03 of 04
Keep each agent small.
One coordinator, a few focused helpers, and clear hand-offs. Easier to test, easier to fix, cheaper to run.
Step 04 of 04
Ship it, then watch it.
Logs, audits, security checks, and test harnesses from day one, so teams can trust it on a busy Monday morning.
Impact
What my work has done so far.
Numbers from real projects at TCS. Tap any card to see where it comes from.
4+ yrs
Production AI engineering
AI Engineer at Tata Consultancy Services since June 2022
Proof · Career timeline50K+
Daily requests served
10+ FastAPI agentic services with sub-second responses
Proof · TCS role100+
Enterprise users on secure LLM proxy
Azure-hosted proxy with tracing and audit, 20% lower latency
Proof · TCS role45%
Manual processing time cut
3+ multi-agent workflows on Microsoft Foundry SDK and MCP
Proof · TCS roleTop 15
Research paper, 10,000+ engineers
Graph-RAG and hybrid retrieval architecture
Proof · TCS role8×
Awards at TCS
Including an On the Spot Award
Proof · AchievementsProjects
A few things I've built.
My favourite open-source work. Star and fork counts update from GitHub every day.
claude architect exam guide
Pinnedsarveshtalele/claude-architect-exam-guide
The CCAR-F, CCAR-P prep guide built for non-technical learners and engineers alike. Curated resources, Claude skills, and a learning path that goes beyond the exam.
AI Projects
18182Updated Sep 2026
personal finance mcp
Pinnedsarveshtalele/personal-finance-mcp
77 deterministic personal-finance calculators + meta-advisor + live market data, served over MCP with a Next.js web UI.
AI ProjectsMCP
Python60Updated Aug 2026
mcp skills registry
Pinnedsarveshtalele/mcp-skills-registry
Self-hostable Model Context Protocol (MCP) server that turns folders of skills & agents into tools any MCP client can discover and run. FastAPI + Next.js, deployable to Hugging Face.
Agentic AIMCP
Python32Updated Oct 2026
devcrew
Pinnedsarveshtalele/devcrew
Turn one AI coding agent into a full engineering team: 15 department subagents, a gated 15-stage SDLC pipeline, and deterministic security, compliance, and token controls. Works in Claude Code, Cursor, and Antigravity. No paid tools.
Claude CodeDeveloper Tools
Shell10Updated Aug 2026
linkedin content skill
Pinnedsarveshtalele/linkedin-content-skill
LinkedIn Content Generator Claude Code Skill
Agentic AIClaude CodeClaude Skills
Python92Updated Jun 2026
ACCIDENT CVPR 2026
Pinnedsarveshtalele/ACCIDENT-CVPR_2026
Zero-shot pipeline for traffic accident detection, localization and classification in CCTV video. Frame-difference peak detection, Farneback optical flow centroids and CLIP prompt matching, with no fine-tuning. ACCIDENT @ CVPR 2026.
Machine Learning
Jupyter Notebook20Updated Sep 2026
Writing
Latest from the blog.
Engineering· 7 min read
I Built a Financial AI That Refuses to Guess
Why I gave an LLM 76 deterministic financial calculators instead of letting it do math, and what that taught me about building AI people can trust.
MCPDeterministic AIFintechAgentic AI
Claude Prep
I passed both Claude exams. Here's how you can too.
I wrote 360 practice questions while preparing for the Foundations exam, and put them all online for free. There's a timed 60-question mock exam, and every answer explains why the other options are wrong.
Toolbox
What I work with.
08
Languages & Frameworks
- Python
- TypeScript
- FastAPI
- LangChain
- LangGraph
- LlamaIndex
- Pandas
- Scikit-learn
09
AI & Agents
- Agentic AI
- Generative AI
- MCP
- Graph-RAG
- Prompt Engineering
- Claude
- OpenAI
- Gemini
- Ollama
06
Cloud & Platform
- Azure
- Microsoft Foundry
- LiteLLM Proxy
- MS SQL Server
- Git & GitHub
- CI/CD
06
Developer Tools
- Claude Code
- GitHub Copilot
- Copilot Studio
- Codex
- Antigravity
- Spec-kit (SDD)
29 tools I use regularly, grouped the way I think about them.
Experience
Where I've worked.
– Present
AI Engineer
Tata Consultancy Services · Thane, India
- Partnered directly with the CXO's office and leadership to architect Agentic AI automations, accelerating Total Contract Value (TCV) growth by 25% and reducing executive workflow bottlenecks by 40%.
- Deployed autonomous workflows to analyse 500+ monthly client meetings and streamline engagement pipelines, increasing follow-up efficiency by 35% and driving data-backed strategic conversions.
–
Industrial Robotics and Applications Trainee
Centre of Excellence in Maritime and Shipbuilding (CEMS) · Vizag, India
- Instructed industrial robotic arms to perform material handling, plasma cutting, and welding.
- Trained and modified a robotic arm to perform design tasks on an industrial robotic cell, delivering realistic KUKA robot handling experience.
–
NX11 and Teamcenter Trainee
Centre of Excellence in Maritime and Shipbuilding (CEMS) · Powai, India
- Completed 3D modeling, assembly, and simulation training in Siemens NX11.
- Used Teamcenter, a Product Lifecycle Management (PLM) system, to analyze a specific product end-to-end.
Let's talk
Have a hard problem? Let's make it simple.
Roles, architecture reviews, Claude certification prep, or a conversation about agentic AI.