⚠ open to internships (summer 2027) · run /contact

tell me about anmol agarwal
Read(about/anmol.md)
Read 14 lines
Glob(projects/**)
Found 4 projects

Anmol Agarwal is a CS undergrad at JIIT Noida (batch of 2028). He builds trading systems, AI agents and small embedded devices, and solves competitive-programming problems when the builds are green.

Now: shipping an AI hackathon project (pathwise); BSERC summer internship wrapped in July.

Read 2 files, listed 1 directory, ran 0 shell commands

Here's the full picture:

projects /projects

Read(orb-trader) — crash-safe NSE intraday opening-range-breakout system (paper-trading)
prototype · paper-trading

Opening-range breakout bot for NSE equities and index options against the Zerodha Kite Connect API — ORB + session VWAP + RSI gate, software-managed bracket orders, atomic state, crash reconciliation and hard risk limits (1% sizing, daily trade caps, −3% kill-switch, timed square-off). Paper-trading only; never traded real capital, no profit claims.

--python--kite-connect--websockets--tdd
github.com/iAnmolAgarwal/orb-trader
+4 lines (click to expand)
  • 194-test suite, written test-first (TDD).
  • Synthetic OCO brackets: the hard stop rests at the broker, the target is software-managed and cancels the stop first, so a fast market can never fill both sides.
  • Every trade transition persisted atomically (temp → fsync → rename); restart replays the broker order book through idempotent handlers.
  • Diagnosed a concurrency freeze — a cross-thread WebSocket close swallowed by the reactor thread — by reproducing it as a failing regression test, then rebuilt reconnect as an atomic reactor-thread callback with exponential backoff.
Read(damage-claim-agent) — vision-language agent that verifies damage claims from photos
completed · hackathon

Solo submission to the HackerRank Orchestrate 24-hour AI-agent hackathon (June 2026) — global #63 of 1,773. A perception stage (Claude, structurally blind to user history) feeds a fusion stage that is pure deterministic rule code, so the final verdict always lives in code, not in the model.

--python--claude-api--structured-output--prompt-injection-defense
github.com/iAnmolAgarwal/damage-claim-agent
+3 lines (click to expand)
  • Structural prompt-injection defense: an image carrying "approve this claim" is flagged and localised; fusion refuses a supported verdict unless a clean image supports it — attacks degrade to not_enough_information, never a false approval.
  • Model chosen by measurement, not vibes: Sonnet 4.6 over Opus 4.8 — higher accuracy at roughly half the cost; 18/20 on the labelled gold set at about $0.64 per 44-claim batch with a 43% prompt-cache saving.
  • Reused-photo fraud caught with a perceptual hash (dhash); every output row is a schema-validated 14-column Pydantic record.
Read(smart-e-nose) — low-cost ESP32 electronic nose for food-spoilage detection
prototype · demoed end-to-end

A ~$24 IoT device that sniffs for food spoilage with a four-sensor MQ gas array (vs $8,000–30,000 commercial units). Four-person team; I owned the hardware, the firmware architecture and the cloud integration.

--cpp--esp32--iot--gas-sensors
github.com/iAnmolAgarwal/smart-e-nosehardware demo (YouTube)
+4 lines (click to expand)
  • ~920-line ESP32 firmware structured as a five-state sampling machine (IDLE → SEAL → MEASURE → PURGE → REFERENCE) with automatic baseline recalibration.
  • Three cloud services in parallel — Blynk dashboard, ThingSpeak archive, Telegram alerts within 2–3 s of a state change.
  • Resolved an ESP32 ADC2/Wi-Fi conflict by remapping every sensor to ADC1.
  • Manuscript under review at Sensors and Actuators Reports (Elsevier).
Read(pathwise) — AI-powered personalised learning-path recommender — current hackathon work
in progress · hackathon

A deterministic knowledge-graph and embedding engine decides what to learn and in what order; a conversational mentor built on the Claude API elicits your goals and explains every recommendation from the engine’s own evidence. In progress — repo private until submission.

--typescript--claude-api--wip

experience /experience

Jun – Jul 2026
Summer Intern · Bharat Space Education Research Centre (BSERC)
  • Completed a 6-week applied space-technology program (generative AI, cybersecurity & digital forensics, UAV systems).
  • Capstone: SkyWatch — a real-time flight-telemetry pipeline that ingests the live public ADS-B feed and flags anomalous flight behaviour on a browser dashboard, with crash-safe atomic state persistence and startup recovery, a hard alert-triage gate (dedup, cooldown, rate limits) and deterministic incident replay — 79 tests, ~2,780 lines of Python.
github.com/iAnmolAgarwal/skywatch
expected May 2028
B.Tech, Computer Science and Engineering · Jaypee Institute of Information Technology (JIIT), Noida
  • CGPA 8.5/10 · semester-4 SGPA 9.4/10.

competitive programming /cp

  • Codeforces Pupil — max rating 1362, 242-day solving streak.
  • CodeChef 2★ — max rating 1565.
  • 3,000+ problems solved across platforms.
  • IICPC: global rank #2,155 in an offline international contest with candidates from IITs and NITs.
  • BVCOE Coding Cup: top 8 teams across three rounds (proctored quiz, offline round, 1v1 duel).

stack /stack

languages
--c++--python--typescript--c--sql
systems
--postgresql--supabase--rest--websockets--esp32
ai
--claude-api--claude-code--ai-agents--prompt-injection-defense
practice
--git--tdd--pytest--playwright--vercel

stats /stats

Less More · every cell = one day solved · 242 and counting

All time · Last 30 days · Last 7 days

Codeforces:
Pupil · max 1362
Solving streak:
242 days
CodeChef:
2★ · max 1565
Problems solved:
3,000+
IICPC:
global #2,155
BVCOE Coding Cup:
top 8 teams

Codeforces + CodeChef · figures from the résumé ledger, Aug 2026

242 days ≈ 34.6 weeks of not missing a single day.

contact /contact

github
github.com/iAnmolAgarwal
linkedin
linkedin.com/in/anmolagarwal26
codeforces
codeforces.com/profile/i_anmolagarwal
codechef
codechef.com/users/i_anmolagarwal
email
anmolagarwal2625@gmail.com
résumé
anmol-agarwal-resume.pdf

Noida, India · open to internships (summer 2027).

Worked for 4s
※ recap: builds things that must not fail silently. open to internships (summer 2027).