Orbit radius is months elapsed since the degree programme began; each body carries a moon for up to six of the technologies it was built with. A planet's hue is its domain, its surface is how far the work got, and a ring marks work that is not solely mine. CivicLens is drawn unfilled because it records no measurements at all. Body size is an ordinal ranking of prominence, not a measured magnitude. The project metrics are not commensurable with one another.
Engineering portfolioUpdated 2026-09-03
I build backend and data systems that keep AI behavior inspectable.
Based in Mumbai, I build backend platforms, streaming pipelines, and applied AI systems with explicit operational boundaries.
Backend Systems / Data Infrastructure / Applied AI
Location
Mumbai, India · Asia/Kolkata
Education
B.Tech in Artificial Intelligence & Data Science, K.J. Somaiya School of Engineering Aug 2023 – Jul 2027 · CGPA 9.71
Most recent
Technology Developer Intern, Barclays June 2026 – August 2026
Orbit radius is months elapsed since the degree programme began; each body carries a moon for up to six of the technologies it was built with. A planet's hue is its domain, its surface is how far the work got, and a ring marks work that is not solely mine. CivicLens is drawn unfilled because it records no measurements at all. Body size is an ordinal ranking of prominence, not a measured magnitude. The project metrics are not commensurable with one another.
§1Experience
Built inside real operating constraints.
Three roles across production software, backend product systems, and analytical infrastructure. Scope notes state what each engagement did and did not cover.
June 2026 – August 2026
Barclays
Technology Developer Intern
Helped move a sanctions-screening platform from proof of concept toward a scalable, configuration-driven system spanning engineering design, processing workflows, and operational tooling.
Participated in system-design proposals with engineering and project stakeholders, translating operational requirements into architecture and workflow.
Built and iteratively evolved a Python, FastAPI, React, and TypeScript platform capable of processing files with 80M+ records across DEV, UAT, and pre-production environments.
Public copy is limited to résumé-level scope and does not imply Barclays endorsement.
September 2025 – November 2025
Makeflow India
Backend Developer Intern
Refactored a Node.js backend for an AI counselling product, separating inference from session state and simplifying a four-stage conversational workflow.
Reduced inter-service coupling by approximately 40% and code redundancy by approximately 30%.
Implemented keyword-triggered crisis routing with a routing decision under 100 ms.
Node.js
JavaScript
REST APIs
Scope: backend routing and product workflow engineering, not clinical decision-making.
December 2024 – January 2025
Segmentriq Analytics
Data Analytics Intern
Built SQL-based ETL and analytics workflows that consolidated operational data into a normalized warehouse and decision-ready Power BI reporting.
Integrated 5+ source systems into a normalized, fact-and-dimension data warehouse.
Delivered 3+ Power BI dashboards with 15+ custom DAX measures for revenue, churn, and funnel analysis.
SQL
ETL
Data Warehousing
Power BI
DAX
§2Selected work
Systems with the evidence left in.
Each system is written up as its own document: the brief, the architecture, the decisions with their costs, what was measured, and where the result stops being true.
§2.1Observability / Distributed Systems
Figure 2.1Shows the incident path from OpenTelemetry input through time and topology scoping to correlation and an evidence-cited diagnosis. The incident identifier and its seven linked signals are an illustrative walkthrough, not a recorded incident, and no latency or accuracy figures appear.
Evidence-grounded incident investigation for distributed systems
An incident-response platform that correlates metrics, logs, traces, topology, and change evidence into auditable root-cause rankings and evidence-cited diagnoses.
Model gate
The deterministic baseline beat the learned ranker, so the model remained ineligible for promotion.
Alternative considered
Replay immutable evidence
Chosen
Store checksum-addressed telemetry bundles in MinIO and perform RCA against replayed artifacts rather than querying live systems during analysis.
Cost
Collection failures and missing signals stay visible; the system does not silently backfill them from a later live state.
Figure 2.2Shows the full flow from public sources through normalisation, the feature engine, a four-way agent fan-out and fan-in, deterministic consensus, evaluation, and the dashboard. The counts it carries (20+ engineered metrics, six dashboard tabs) describe the system's shape. No measured scores, timings, or throughput appear.
A research-oriented Indian mutual-fund ranking system that combines public data, role-specialised agents, deterministic consensus behavior, and evaluation tooling.
Evaluation finding
The persisted study found a directional, but not statistically significant, advantage over a simple baseline.
Alternative considered
Separate evidence from model judgement
Chosen
Engineer return, risk, cost, and consistency objectives before asking four role-specialised agents to rank candidates.
Cost
The role prompts are deliberately prescriptive, so the agents are closer to specialised rerankers than unconstrained analysts.
Figure 2.3Shows the Kafka to Spark to inference path together with the benchmark's own measured values: 3,285 events per second, a 26.53 ms median batch, 99.2% AUC retained on a reduced feature set, 284K+ transactions, and 577:1 class imbalance. Every figure comes from a simulated local streaming benchmark, not production traffic.
An interactive analysis of Indian IPOs from 2019–2024 spanning enriched returns, regression, clustering, distribution analysis, and sector-level exploration.
I work across the seams: APIs expose workflows, data preserves evidence, and models are tested against explicit baselines before anything depends on them.
Systems
Typed APIs, authentication, failure handling, and operational workflows with explicit boundaries.
FastAPI
Spring Boot
Node.js
PostgreSQL
Redis
Data
Batch and streaming pipelines designed around workload, lineage, replay, and measurable behavior.
Kafka
Spark
PySpark
Pandas
Warehousing
Intelligence
ML and LLM workflows evaluated against baselines before they are trusted inside a product.
PyTorch
TensorFlow
MLflow
Spark MLlib
LLM systems
Where the domains meet
Systems→Kafka→Data
Carries durable event streams between services and data processing.
Systems→FastAPI→Product
Exposes backend workflows through typed, product-facing APIs.
Data→PyTorch→Intelligence
Turns prepared data into trainable and evaluable model behavior.
Systems→PostgreSQL→Data
Provides transactional storage for data-intensive services.
§4Background
The model is only one part of the system.
I gravitate toward systems where a model is only one component: evidence has to be collected, decisions replayed, and failure boundaries made explicit. That is why my projects pair AI experiments with baselines, tests, and visible limitations.
My work spans backend services, data infrastructure, and applied AI. Not as isolated demos, but as systems that can be tested, reviewed, and understood by someone who did not build them.
Institution
K.J. Somaiya School of Engineering
Programme
B.Tech in Artificial Intelligence & Data Science
Period
Aug 2023 – Jul 2027 · In progress
CGPA
9.71
Building a backend, data, or applied-AI system?
If the hard part is making it trustworthy, operable, or understandable, I would be glad to compare notes.
