Sahil Chakraborty
Dimension

AI Engineer · Kolkata, India · open to Bangalore / Delhi / Remote

Ships AI agents, not notebooks.

Data engineer turned AI engineer who ships agents with guardrails, hard cost ceilings, and published numbers — including the ugly ones.

View the workGet in touch

Goals · Aug 2026

Where this is heading.

Next three months

Shipping Yardstick, interviewing, a published fine-tune, deeper document-RAG, and picking up a new stack.

In twelve months

Building higher-value AI systems — bigger scope, greater impact, steeper learning.

Work

Shipped systems, measured — including the ugly numbers.

DataChat

Agentic natural-language analytics over public datasets — ask in plain English, get safe verified SQL and a grounded answer.

Execution accuracy
0.810
Refusal accuracy
1.00
SQL valid rate
0.952
LangGraphFastAPIPydantic v2SQLAlchemy 2 (async)Postgres + pgvectorRedis
Details →CodeLive demo

Quorum

A supervisor agent that reviews pull requests, grounds every finding in the target repo's own docs, and won't post anything without a human approving it.

Retrieval NDCG@5
0.526
Reranking delta
-0.079 at 63-91x latency, cut
AST context-scoping
-34.86% tokens
PythonFastAPILangGraphMCP (client + server)Postgres + pgvectorHybrid retrieval (dense + BM25)
Details →Code

ITR-6 Text-to-SQL

Plain-English querying of Indian corporate tax filings — where the real engineering is the safety layer that stops a model inventing a column and running it.

Statutory line items
129
Independent safety gates
2
Tests
55
PythonSQLiteGroqStreamlit
Details →CodeLive demo

Electricity Production Forecasting

Forecasting 33 years of monthly US utility production — with a naive baseline kept in the table, so a model has to earn its complexity.

SARIMA (best model)
MASE 0.959
Holt-Winters
MASE 1.308
Dataset
397 monthly obs
PythonstatsmodelsSARIMAHolt-WintersStreamlitPlotly
Details →CodeLive demo

Breast Cancer Classification

early

A classic supervised-classification project on a standard diagnostic dataset — an early ML fundamentals build.

scikit-learnLogistic RegressionRandom ForestpandasMatplotlib / Seaborn
Details →Code

Currently building · since Aug 2026 — Yardstick. An agent that trains candidate models on tabular data and reports honestly how good they actually are: automated leakage detection, every model compared against a trivial baseline, and metric selection that is justified rather than defaulted.

Stack

Grouped by what I'd be trusted with.

Flip a card for what it is and which projects I used it in — not an alphabetised keyword dump.

Interviewable — grill me on these

Familiar with — used, not claiming depth

LangChaincross-encoder rerankingRAGASLangSmith / LangfusePyTorchHugging Face TransformersChromaDBFAISSMySQLJenkins CI/CDLinux / WSL2Azurellama.cpp

Experience

From data engineering into AI/ML.

Outcomes and ownership — not a list of responsibilities.

  1. Platform Engineer cum Data Engineer

    Jul 2025 – Present

    Tata Consultancy Services

    • Ship features inside an enterprise GenAI/LLM framework — prompt orchestration and LLM response handling — and version production prompt templates against held-out tests.
    • Build Python and SQL ETL with automated integration and ingestion across the billing domain, on AWS (S3, EC2, IAM).
    • Profile source systems for lineage, quality, and PII exposure — the ingestion and redaction layer a RAG corpus needs.

  2. AI/ML Intern

    Dec 2023 – Jan 2024

    LTIMindtree

    • Fine-tuned FLAN-T5 with PEFT for a text-to-SQL task over ITR-6 filings; built the training corpus end to end — extraction, cleaning, entity tagging.

  3. B.Tech, Computer Science & Engineering

    2021 – 2025

    Kalinga Institute of Industrial Technology (KIIT), Bhubaneswar

    CGPA 8.89 / 10

Certifications

Verifiable, dated, real.

Microsoft Certified: Fabric Data Engineer Associate

Microsoft · Feb 2026

ID E99A9058AB1A4B93

Microsoft Certified: Azure Data Fundamentals

Microsoft · Dec 2025

ID CD5636B3165D6A2

Generative AI: Introduction and Applications

IBM · Aug 2024

ID 2HLWPEDNGHQQVerify →

Generative AI: Prompt Engineering Basics

IBM · Aug 2024

ID 08IQAHT9UAXNVerify →

Prompt Engineering for ChatGPT

Vanderbilt University · Aug 2024

ID RJGE4F3649FJVerify →

AI For India 2.0

HCL GUVI · Aug 2023

ID B22904Du9WrQ0d1496

Python

HCL GUVI · Aug 2023

ID B5173G6i8m610SP9U2

Off the clock

Three things I actually do.

Riding

The peace I feel — bikes mostly, sometimes the car.

Gaming

Ex-esports on PC — Apex Legends (#4, Asia server) and Valorant.

Finding new tech

On the frontier — quantization, transformers, whatever's next.

Contact

Let's talk.

Messages come straight to my inbox and I reply myself.