Data, ML and AI engineer, Boston.
M.S., Northeastern, Dec 2026.

Santhoshkumar Chandrasekar
Quick look Whole building Built from
Reach out Résumé

Santhoshkumar Chandrasekar

Santhoshkumar Chandrasekar, smiling, in a black sweater

A building that wasn’t
built in a day

What I learned

“The most personal is the most creative,” a quote by Martin Scorsese, is one I can attest to with my life.

With advancements blooming in every corner of the world through AI, I feel we are no longer at “This hasn’t been done.” We should be moving towards “This hasn’t been done THIS WAY.” There are millions of ways, unique ones, just flying around for us to grab, utilize, and cultivate, and I am more ready than ever before to build those ways with a team that wants them!

The next floor is still under construction.

I graduate in December 2026 and I’m looking for a full‑time role in data analytics, machine learning or AI. If you have one in mind, or just want to talk about the work, write to me.

Write to me

Built with three.js. Moon, Earth and stars from NASA (public domain); city panoramas, textures and models from Poly Haven (CC0); type from Google Fonts (SIL OFL).

Santhoshkumar Chandrasekar

Data analytics, machine learning and AI engineer in Boston. M.S. Data Analytics Engineering at Northeastern, graduating December 2026.

Anna University

Education, College of Engineering, Chennai, Aug 2017 – Aug 2021. B.E., Electronics & Communication.

Four years at the College of Engineering, Guindy. Mostly signals and systems: how to pull a clean signal out of a noisy measurement. I didn’t know yet that I’d spend the next five years doing the same thing to spreadsheets.

What I did

  • Electronics and communication engineering, with the emphasis on signal processing.

LTIMindtree

Experience, Chennai, 2021 – 2022. Software Engineer.

First job out of college. An insurance client whose policies, claims, agents and premiums each lived in their own system, and a lot of data that didn’t agree with itself.

75% more efficient data flow; 30% less redundant data; 20+ client dashboards.

What I did

  • Rebuilt the ETL that joined the client’s systems, in Informatica, SSIS and Azure Data Factory: 75% more efficient data flow, 30% less redundant data.
  • Ran it in production through CI/CD and Git: watching the overnight jobs, fixing the ones that failed.
  • Built the SSAS OLAP cubes and dimensional models underneath, and 20-odd Power BI and Tableau dashboards on top.

Looking back

The overnight jobs taught me more than the build did. A pipeline that nobody watches isn’t finished.

Barclays

Experience, Chennai, 2022 – 2024. Data Analyst, Mortgage Product Owner.

From building pipelines to owning a product. The question on my desk was simple to say: which mortgage customers were worth reaching, and which were being bothered for nothing?

+40% campaign response; −70% reporting turnaround; +15% user retention.

What I did

  • Product Owner for the mortgage portfolio, between the data scientists and the campaign managers.
  • Segmented customers on transaction and engagement data and rebuilt the campaigns around the segments: +40% response, 5% less wasted marketing spend.
  • Ran 50+ campaigns on Teradata and the CRM, tracked on Tableau and Excel dashboards; user retention up 15%.
  • Replaced the hand-built weekly reporting with SQL and Python pipelines, cutting turnaround by 70%.

Looking back

A segment the campaign team doesn’t trust is a spreadsheet. I spent as much time explaining the models in their terms as building them.

Northeastern

Education, Boston, 2024 – 2026. M.S., Data Analytics Engineering.

In 2024 I moved from Chennai to Boston. I arrived having touched machine learning; I’m leaving as someone who builds with it: pipelines, models, release gates, and the three projects on the floors above. 4.0 GPA; graduating December 2026.

4.0 GPA, out of 4.0; Dec 2026 graduating.

What I did

  • Coursework across data management, storytelling, mining, MLOps, deep learning and GenAI.
  • Built MOMENTO in MLOps (IE7374), which placed 2nd of 30 at a Google-sponsored expo, THE EYE in Deep Learning (IE7615), and Bat-Terry in Applied Generative AI (IE5374); all three are floors of their own.

Fidelity Investments

Experience, Fidelity Investments, Boston, Jul – Dec 2025. Data Analyst Co-op.

A six-month co-op. A new Health & Welfare offering had landed on the team with plenty of data and no way to read it: siloed legacy systems, each with its own answer to how many employees we covered, and no agreed metrics, reporting framework or data dictionary.

1.3M+ employees unified; 3M+ enrollments; 12M+ engagements a year.

What I did

  • Wrote the metric definitions first, then the reporting framework and data dictionary behind them: the first ones the offering had.
  • Unified the legacy systems into one Snowflake source of truth, with accuracy, completeness and consistency checks on every load: 1.3M+ employees, 3M+ enrollments, 12M+ engagements a year.
  • Built the Tableau Factbook on top: correlation, key-driver and trend analysis across 50+ plan sponsors.
  • Walked it through with Sales, Marketing, Health and senior leadership, each of whom needed a different page.

Looking back

Two teams had each been reporting their own version of the same number, and both were defensible. Until there was one definition, nothing we published could be argued with, so that came first.

MOMENTO

Project, Northeastern, with my team, Jan – Apr 2026. Reader compatibility platform, MLOps.

Two people can give the same book five stars and have read completely different books. MOMENTO matches readers by how they think and feel about the same passage, not by what they rated it. The idea, the ML design and the front end were mine; the team made the rest real.

2nd of 30 teams, Google-sponsored MLOps Expo; 450 interpretation benchmark.

What I did

  • Compatibility engine: a Gemini 2.5 Flash agent scoring each reading on the Think and Feel dimensions, and a Bradley-Terry ranking model that learns from engagement.
  • A three-metric release gate backed by an offline evaluation harness and a 450-interpretation benchmark.
  • PII, injection, profanity and spam filters, with retraining triggered when the metrics regress.
  • 2nd of 30 teams at a Google-sponsored MLOps Expo.

Looking back

The release gate caught regressions that a demo would have hidden. That is the whole argument for building the harness before the feature.

Try the demo · See the code

THE EYE

Project, Deep Learning, Northeastern, Sep – Oct 2026. Explainable object detection.

A detector for 73 everyday objects that explains every mistake it makes: wrong ID, missed, or extra. It started with 6,942 unlabeled photos and a question most detection projects skip: when the model is wrong, why?

0.971 test mAP@0.5; 0.76 → 0.92 recall on rotated grids; 0.37 → 0.88 recall on cut-out objects.

What I did

  • Labeling with a human in the loop: OWLv2 prompts and a self-trained YOLO voted on every box, and the disagreements went to review. 98.9% label accuracy.
  • Diagnosed its blind spots and retrained on them: recall rose from 0.76 to 0.92 on rotated grids and from 0.37 to 0.88 on objects cut out onto new backgrounds, and test mAP@0.5 reached 0.971.
  • Four detectors and a classifier compared side by side, with the selection rule written and SHA-256-sealed before any result came in, and every number reported with a bootstrap confidence interval.
  • Deployed on Oracle Cloud with HTTPS, auto-restart and per-user rate limiting.

Looking back

COMPOUND, an ensemble of all five models, got 6.9 points more images fully correct. It also lost recall on cut-out scenes and invented boxes, and the sealed rule said no. I kept the model the evidence supported, not the one with the better headline.

Try it live · See the code

Bat-Terry

Project, Applied Generative AI, Northeastern, with my team, Nov 2026. Goal-inversion planner for Boston’s 311 crews.

A bat called Terry who plans backwards. Name a goal for Boston’s 311 crews, and Terry imagines the year it went wrong, tests every reason against real 311 records, and finds the cheapest crew plan that avoids them. I originated the idea and built it with my team, who helped with testing and deployment.

98.8% right department, English or Spanish; 0 invented request types in 24 flips; 90%+ on time, 14% under to 14% over forecast.

What I did

  • A hackathon brief: staff Boston’s 311 crews for 2025. In 2024 the city took 306,756 requests, and Transportation closed only half of them on time.
  • The flip: Terry turns a goal into the year it failed and names its predators, which must be real request types and neighborhoods, each tested against 2024 records. Across 24 flips, no invented request types, and 56% of testable predators confirmed.
  • The flyway: a forecast, a day-level crew simulator and an optimizer find the cheapest crew plan that avoids the predators, priced across a quiet, a normal and a busy year. The Neutral plan stays at least 90% on time from 14% under to 14% over forecast.
  • Echolocation: an equity map of Boston, showing who calls 311 less than expected and who waits longer.
  • An intake assistant routes residents’ messages, in English or Spanish, to the city’s real request types: 98.8% right department. Ping Terry answers questions from the plan’s own numbers, and 42 of 42 answers passed the number check.

Looking back

A plan built only for the expected year broke at 5% more demand. So Terry plans for three possible years, not one.

Try it live · See the code · Watch the film

What I’m built from

  • Creativity is my concrete. I was built on drawing, scriptwriting, coding and SQL’ing problems into unique solutions, so everything I work on has creativity cemented into it.
  • The urge to do something new is my steel. It’s what keeps me rising past what’s already been done.
  • Transparency is my glass. I’m upfront about what I expect and where my progress stands, because trust and boundaries are what let a team shine through.
  • Curiosity is my wiring. It took me from circuit boards to neural networks, and it has lit up every floor I’ve stepped into.
  • My family and friends are my mortar. When the ground shifted, they held my bricks together.