BACKEND & SYSTEMS ENGINEERING · DEVELOPER TOOLS · RELIABILITY · OPEN SOURCE
Integrated M.Tech candidate in Mathematics & Computing at IIT (ISM) Dhanbad · Expected graduation: May 2027
| Project | What it demonstrates | Stack |
|---|---|---|
| Low-Latency Matching Engine | Price-time order matching, cancellation, bounded event path, benchmark methodology | C++20 · CMake |
| RepoPilot | Local code indexing, evidence-grounded search, API and dashboard | Python · SQLite · React |
| FluxQueue | Durable jobs, leases, idempotency, retries and dead-letter handling | FastAPI · PostgreSQL |
| Uptime Sentinel | Scheduled checks, incidents, worker coordination and status pages | FastAPI · PostgreSQL |
| StoreForge | Content-addressed object storage, replicated chunks, immutable versions, repair and recovery workflows | Python · FastAPI · PostgreSQL · React |
| UPI Offline Mesh | Local-only offline-payment simulation with encrypted QR payloads and idempotent settlement; API authentication remains unfinished | Java · Spring Boot |
| DataPulse Analytics | Event ingestion, aggregation, migrations and dashboard | FastAPI · PostgreSQL · React |
|
Define state transitions, ownership rules, bounds, and idempotency guarantees first. Keep the matching core deterministic and make APIs reject invalid states at their boundaries. |
Test retries, timeouts, stale leases, partial failure, duplicate requests, recovery, and cleanup—not only successful requests. Fail closed when safe behavior cannot be established. |
|
Prefer source-linked evidence, reproducible tests, versioned data, migration checks, and explicit abstention over claims that cannot be verified. A passing metric must always state its scope. |
Treat repository content and external inputs as untrusted. Validate paths and payloads, protect credentials, constrain response size and execution time, and keep security boundaries explicit. |
|
Benchmark real workloads, record environment and compiler settings, inspect distributions and run-to-run variance, and distinguish a local demonstration from production guarantees. |
|
DESIGN → TEST → OBSERVE → IMPROVE
SYSTEMS · DEVELOPER TOOLS · JOB PROCESSING · RELIABILITY · PAYMENTS · ANALYTICS
A C++20 limit-order-book prototype with price-time priority, full and partial fills, indexed cancellation, order-ID lookup, and a bounded single-producer/single-consumer event path. The matching core is deliberately single-threaded; concurrency is isolated at the event boundary.
STACK C++20 · CMake · SPSC queue · Sanitizers · Benchmarks
VERIFICATION The repository's CI includes Release tests, a Debug AddressSanitizer/UndefinedBehaviorSanitizer run, cancellation-heavy workloads, and multi-workload benchmark artifacts. Throughput and latency depend on the measured environment; the README documents the workload and interpretation limits.
VIEW SOURCE ↗ · CI & artifacts ↗ · BENCHMARK METHODOLOGY ↗
RepoPilot indexes local codebases and combines incremental indexing, Python symbol extraction, conservative cross-language import/include edges, SQLite FTS5 retrieval, line-addressable evidence, an API, optional MCP tools, and a React dashboard.
STACK Python · SQLite/FTS5 · FastAPI · MCP · React · TypeScript
VERIFICATION CI exercises Python 3.11/3.12, the offline single- and multi-repository evaluation suites, repository performance benchmarks, Windows file-safety tests, frontend browser tests, Docker/authentication smoke tests, CodeQL, and the optional MCP import. The 41-case evaluation still needs independent second-person review (tracked in RepoPilot issue #33); structural citation checks do not prove semantic entailment. Live-provider evaluation is optional and separate from ordinary CI.
VIEW SOURCE ↗ · CI & EVALUATION ARTIFACTS ↗ · EVALUATION LIMITS ↗
A durable job-processing platform built around validated job submission, tenant-scoped API keys, idempotency, PostgreSQL persistence, leases and heartbeats, stale-token fencing, bounded retries, terminal dead-letter state, and payload-free worker logs.
STACK Python · FastAPI · SQLAlchemy · PostgreSQL · Alembic · Docker
VERIFICATION Main CI and the documented manual smoke workflow cover database migrations, worker execution, lease/concurrency behavior, API authentication, and container startup. Merged PR #20 bounds handler waiting to 30 seconds by default and limits unresolved handler threads to eight per worker process; a timed-out thread may continue running until it exits, so side-effecting handlers must be idempotent. The PR's lint, formatting, unit/API, PostgreSQL concurrency, dependency-audit, CodeQL, and container checks passed. The free-tier hosted endpoint may sleep and has no production availability guarantee. Treat the demo as a portfolio demonstration, not an uptime commitment.
VIEW SOURCE ↗ · CI RUNS ↗ · API DEMO ↗
A FastAPI service for scheduled HTTP/HTTPS checks, latency and failure history, incident open/update/recovery, a read-only public status page, and a PostgreSQL-backed worker using shared leases and fencing.
STACK Python · FastAPI · PostgreSQL · SQLAlchemy · Docker Compose
VERIFICATION PRs #16, #18, and #19 are merged. PR #16 fences expired worker leases before result persistence or rescheduling. PR #18 moves blocking DNS validation off the event loop, bounds it by the monitor-check deadline and a five-second ceiling, and limits concurrent resolver threads. PR #19 applies bounded DNS validation to monitor creation and URL updates, returning 503 on resolver-capacity exhaustion and 504 on timeout. The recorded CI for #18 and #19 passed lint, formatting, typing, tests, PostgreSQL concurrency, migrations, local proxy security integration, production Compose validation, and Compose smoke checks. The optional external controlled-proxy suite remains skipped; production egress security is not certified. Paid hosting and production readiness are out of scope.
VIEW SOURCE ↗ · CI RUNS ↗ · VERIFICATION REPORT ↗ · RELEASE CHECKLIST ↗
An offline-payment prototype exploring encrypted QR payloads, AES-GCM authenticated encryption, RSA-PSS signatures, duplicate/replay protection, and idempotent settlement once a message reaches the simulated gateway.
STACK Java 17 · Spring Boot · Cryptography · PostgreSQL · Flyway
SCOPE & VERIFICATION The repository implements a backend settlement gateway, QR payment simulation, and server-side store-and-forward/mesh simulation. Native Bluetooth LE/Wi-Fi Direct transport, a native Android client, and real payment rails are not implemented. The current REST API lacks user authentication, authorization, and authenticated bridge identity; the project is restricted to isolated local/test use and must not be exposed publicly. Remediation is tracked in security issue #15. Java CI includes persistence and production-profile tests; these validate a prototype, not production payment security.
VIEW SOURCE ↗ · CI RUNS ↗ · DEMO VIDEO ↗
A portfolio-scale event analytics demo: validated event ingestion, SQLAlchemy persistence, metrics/top-event aggregation, PostgreSQL migrations, and a React + TypeScript dashboard.
STACK Python · FastAPI · Pydantic · SQLAlchemy · PostgreSQL · React · TypeScript
VERIFICATION CI runs backend endpoint and payload-validation tests, applies the initial Alembic schema to PostgreSQL, checks migration/model parity, and verifies that production startup does not silently create the schema. Full user authorization, ingestion rate limiting, and broader frontend/API workflows remain future work.
MORE BUILDS — ETL · TIME SERIES · FRONTEND
A standard-library Python ETL pipeline for synthetic healthcare encounter events: input validation, normalization, duplicate handling, deterministic aggregation, and quality reports with row-level rejection reasons.
VERIFICATION CI checks the unit suite and sample end-to-end pipeline. Regression tests cover Decimal rounding, malformed CSV, duplicate headers, surplus fields, invalid dates, and clear input-row diagnostics. All data is fabricated; this is not a clinical or billing system.
A Python object-storage prototype with PostgreSQL metadata, three-node replicated chunk storage, immutable object versions, durable repair jobs, per-tenant token isolation, a React dashboard, and backup/restore tooling. The Compose nodes are not independent hardware failure domains. Production deployment remains gated on independent tenant-isolation/security review, deployment TLS/firewall/shared rate limiting, off-host recovery drills, and environment-specific benchmarks. See release-readiness issue #12.
A time-series forecasting exploration for Indian-city AQI with strict CSV validation, chronological persistence and training-mean baselines, and aggregate/per-series evaluation metrics. The combined evaluation and README update is merged in PR #8; its tests, baseline runs, notebook checks, dataset checks, and metrics-artifact upload passed CI. Historical ARIMA/SARIMAX/Prophet metrics remain unverified because their original evaluation protocol was not fully documented.
A responsive React task manager with accessible controls, active/completed filters, input validation, empty states, and browser-local persistence. It is intentionally a frontend learning project, without a backend or cross-device sync.
A separate 3D activity visualization, refreshed by GitHub Actions. GitHub’s native contribution graph below the README remains the authoritative yearly total.
CPython #160707 — MERGED
Free-threaded garbage-collector traversal fix and regression coverage. · Review the contribution ↗
CPython #160703 — CLOSED · NOT MERGED
Monitoring callback error propagation from for loops, with a regression test. Tier-2 runtime coverage was not demonstrated in the available build; the proposal was closed without merge. · View contribution ↗
chunklet-py #107 — MERGED
Windows multiprocessing pickling fix and regression tests. · Review the contribution ↗
jvm-mcp #238 — MERGED
Spring bean discovery failures are reported as errors instead of misleading empty results, with safer diagnostics and regression coverage. · Review merged contribution ↗
jvm-mcp #239 — MERGED
Detect Java versions from target JVM executable paths, with implementation and regression coverage merged upstream. · Review merged contribution ↗
jvm-mcp #240 — MERGED
Preserve original JMX connection failure causes and close partially initialized connectors safely, including cleanup-failure tests. Timeout support remains explicitly out of scope. · Review merged contribution ↗
jvm-mcp #241 — MERGED
Make virtual-thread inspection opt-in, bound temporary thread-dump size, and test cleanup and Java 21 behavior. · Review merged contribution ↗
jvm-mcp #242 — MERGED
Reduce false-positive framework detection and resolve relative JAR paths against the target JVM working directory on Linux when available. Windows/macOS resolution remains limited. · Review merged contribution ↗
ATTRIBUTION NOTE
jvm-mcp #230 belongs to another contributor (Chandu0501) and is merged upstream; it is not listed as your contribution. Your own jvm-mcp PRs #238–#242 are merged; CPython #160714 remains your open proposal. The separate #230 PR is not your contribution.
CPython #160714 — OPEN
kqueue interruption-handling proposal and regression test. · View open proposal ↗
Merged contributions and open proposals are intentionally distinguished.
EXPLORE ALL REPOSITORIES ↗ · PROFILE OVERVIEW ↗
Live panels use external services and may occasionally be delayed or unavailable. The project illustrations above are design visuals, not live application metrics.


