Resume Validation | DSA Grind

Audience: Sanjay (3.5 YoE, current title Senior SWE @ Jio). Target roles: MAANG SDE-II / L4 / E4 equivalent, and Healthtech-AI senior IC. Style of this review: blunt. No flattery. Every critique is anchored in a direct quote from your current resume.

1. Overall verdict

Score: 6.8 / 10 for MAANG SDE-II target. 7.5 / 10 for Healthtech-AI senior IC.

You are not a “throw this resume in the trash” candidate. You’re a “this resume buries a real candidate under buzzwords” candidate. The underlying work is legitimately above-average for 3.5 YoE — CDC with Debezium in prod, an LLM rate limiter, FHIR/OpenEHR, a multithreaded reporting engine across 80+ hospitals. That’s genuine SDE-II ammunition.

What’s blocking you from a 9/10:

Tone. The resume reads like a manager wrote it about himself. MAANG screeners (especially Google, Meta) are allergic to words like “Strategic,” “Architecting,” and capitalised abstract nouns mid-sentence.
No public proof. Zero LeetCode link, zero blog, no pinned GitHub repos referenced, no system-design writeup. For MAANG SDE-II, this is the single biggest gap.
Unbacked claims. “Millions of users” in summary is never substantiated in any bullet. Recruiters notice this in 8 seconds.
Project section drags you backwards. Octave music app and image colorization are fresher projects. They lower the perceived seniority of the whole document.

You’ll get phone screens at most MAANG companies on the strength of the GenAI + healthcare + CDC story. You’ll struggle in the resume-deep-dive part of the loop unless you fix items 1–4.

Realistic outcome with current resume (no other prep):

MAANG SDE-II phone screen: ~40% callback from cold apply, ~70% via referral.
Healthtech-AI senior IC (Tempus, Abridge, Komodo, Innovaccer, Nuance/Microsoft Health): ~65% cold, ~85% referral. This is your strongest market.
Onsite conversion: capped at ~25% until you add public proof + DSA evidence.

2. What’s STRONG (and exactly why)

#	Strength	Why it works
1	GenAI in production at 3.5 YoE — “AI Rate Limiter — 40% LLM cost reduction” and “Clinical AI Summarization Engine”	Most candidates at 3.5 YoE have toyed with LLMs. You’ve shipped cost-control infra for them. This is a 2025–2026 differentiator.
2	CDC pipeline with Debezium + Kafka, sub-3s	Hard, prod-grade distributed-systems work. Interviewers can drill on this for an entire system-design round.
3	Regulated-domain depth (HIPAA, FHIR, OpenEHR)	A real moat for healthtech-AI. Most generalist SDE-IIs cannot pass a FHIR conversation.
4	Quantified outcomes throughout — 60%, 40%, 80%, 70%, 50%, 8x	Numbers exist on most bullets, which beats 80% of resumes at your level. The credibility of the numbers is a separate question (see Section 3).
5	Multithreading + iText reporting engine across 80+ hospitals, 21+ encounter types	Concrete scale + cross-customer impact. Behavioral-round gold.
6	Hibernate 2.5s → <500ms (80%)	Classic JPA-tuning story, low-risk to defend, easy STAR narrative.
7	Cross-functional / mentorship signal — “Technical Governance + KT sessions → 40% sprint velocity gain”	MAANG looks for “scope of impact beyond your code” for L4/E4. You have evidence.
8	Two clean promotions — Intern → SWE → Senior SWE inside ~3 years at one org	Promo trajectory itself is a positive signal; companies infer “this person did not coast.”

Keep all of these. The fixes below are about polish, not amputation.

3. What’s WEAK or RISKY

Critiques are quoted verbatim from your resume so there’s no ambiguity.

3.1 The summary is a buzzword bath

“Strategic Senior Software Engineer with over 3.5 years of experience in architecting high-concurrency microservices and integrating Generative AI to optimize the SDLC.”

Three problems in one sentence:

“Strategic” is a manager word. ICs at MAANG do not call themselves strategic. Drop.
“architecting high-concurrency microservices” — at what QPS? 50? 50k? Without a number this is filler.
“optimize the SDLC” — vague. SDLC is a process, not a system. Recruiters cannot picture what you actually did.

3.2 Unsubstantiated “millions of users”

“…cost-optimizing LLM orchestration for millions of users.”

Search your own resume — no bullet anchors “millions of users.” Either:

Add a bullet with the concrete MAU number, or
Drop the phrase. An interviewer who catches an unbacked claim will be 30% more skeptical of every other number.

3.3 Capitalised abstract nouns mid-sentence

“automated Architectural Compliance + dependency management” “HA maintained” “Scalability”

Capitalising “Architectural Compliance” and “Scalability” looks like you copy-pasted from an LLM and didn’t lowercase. Two effects: (a) recruiters smell AI-generated copy and (b) it reads as marketing, not engineering. Lowercase these.

3.4 Skills section is bloated

“Languages: Java, Python, SQL, JavaScript, TypeScript, C++”

C++ — do you use it weekly? If no, remove. A staff interviewer who sees C++ on a Java/Python resume will throw a C++ question. Don’t bait it.
JavaScript + TypeScript both — pick one, or move to “Familiar.” Listing both implies depth you probably can’t defend.
Django — appears in skills, but no bullet uses it. FastAPI is mentioned in bullets, Django isn’t. Drop Django.
“Postman Automation” — not a skill, it’s a tool you used. Move into a tools line, don’t headline it.
“Factory Design Pattern” as a skill — design patterns aren’t a skill, they’re table stakes. Drop the explicit mention; show it through system-design writeups instead.

3.5 Numbers that interviewers will drill

Claim	What an L5 interviewer asks	Risk if unprepared
“60% dev-time reduction” via AI-Driven Dev Framework	“How was baseline measured? Sample size? Confounders?”	High — sounds like a vanity metric.
“40% LLM cost reduction”	“Over what time window? Absolute $/mo? Did request volume stay constant?”	High — need absolute number, not just %.
“200+ SonarQube vulnerabilities mitigated → HIPAA compliance”	“How many were Critical/High vs Info? Did you actually achieve HIPAA certification or just mitigate?”	Medium — HIPAA is a process, not a SonarQube outcome. Soften the link.
“50% manual QA reduction” via auto test collections	“Hours saved per sprint? Did defect-escape rate stay flat?”	Low if you have the math, high if you don’t.
“8x speedup via FFmpeg + GPU”	“Baseline machine, codec, resolution, batch size?”	Medium — easy to defend if you owned it, embarrassing if you didn’t.

Rule: for every % on the resume, have the absolute number and the measurement method ready in your head. If you can’t recall one, soften the claim now.

3.6 Intern section is too long

Seven bullets for an internship in 2022–2023, two years ago, is too many. At Senior SWE level, an internship should be 2–3 bullets max — just enough to prove “I shipped before I was paid full-time.” Compress.

3.7 Project section drags seniority down

“Octave music web app (ReactJS, Firebase) — 1000+ users” “Black & White Image Colorization (Python, CNN, GAN, TensorFlow, OpenCV) — 85% color accuracy”

These are good undergraduate projects. They are bad on a Senior SWE resume because:

They date back to before your full-time experience.
They imply you don’t have a more recent personal project worth listing.
They cannibalize the prime mid-page real estate that should be amplifying your healthcare-AI moat.

Keep at most one (the video transcoding one is fine — Java 21, Kafka, 8x speedup is on-brand). Drop the rest, or replace with a 2025 side project.

3.8 Web presence is invisible

“LinkedIn | github.com/snaik4398 | sanjaydev.online”

Three URLs listed, zero context for two of them. A recruiter will not click sanjaydev.online unless you tell them what’s there. Either:

Make sanjaydev.online a 1-page portfolio with 3 case studies, OR
Drop it.

GitHub link is in the contact line but no pinned repos are referenced anywhere on the resume. Recruiters do not browse your repos uninvited. Reference 1–2 by name in a “Selected Repos” line.

3.9 Subtle red flags

“Strategic” in summary — drop.
“Proven track record” — cliché phrase, hurts on ATS-aware reviewers.
“Expert in…” — at 3.5 YoE, “Expert” reads as overclaim to anyone L6+.

4. Gap analysis for MAANG-specific signals

Signal MAANG looks for	Sanjay has it?	If not, how to get it in 3 months
DSA / problem-solving public proof	NO — no LeetCode profile listed	Solve 150 LC problems, make profile public, add URL to contact line
System design writeups	NO	Publish 1–2 short blog posts (Medium / dev.to / personal site): the LLM Rate Limiter and the Debezium CDC pipeline are perfect topics
Open-source contribution	UNCLEAR — github.com/snaik4398 listed, no specific PR cited	If you have ≥1 merged OSS PR, cite it by repo + PR number. If not, leave alone — don’t fabricate.
Scale numbers (QPS, users, data volume)	PARTIAL — 80+ hospitals is good, but no QPS, no MAU, no GB/day	Backfill bullets: peak QPS of the reporting engine, GB/day through Debezium, request rate at the rate limiter
Mentorship / leadership at IC level	YES — “Technical Governance + KT sessions → 40% sprint velocity”	Keep. This is already L4/E4-grade signal.
GenAI / LLM productionization	YES — strong	Lean harder for AI-first companies. Lead with the rate-limiter story.
Healthcare / regulated-domain expertise	YES — FHIR, OpenEHR, HIPAA	Strong moat for healthtech. Downplay (don’t remove) for Meta/Google generalist apps.
Cross-functional impact	YES — 80+ hospitals, 5 distributed services, 20 microservices LiquiBase	Good. Surface this higher in the experience block.
Observability / SRE signal	NO — no mention of SLO, SLI, p99 latency, dashboards, alerting	Add one bullet: “Defined SLOs for X, drove p99 from Y to Z via …” if true
TDD / testing discipline	PARTIAL — Postman & JUnit mentioned	Add coverage % or test-pyramid bullet if you have one
Production incident response	NOT VISIBLE	If you’ve owned an on-call rotation, add it. Recruiters at L4+ ask.

5. Rewrites — 5 weakest bullets, before / after

Rewrite 1 — AI-Driven Dev Framework

BEFORE: “AI-Driven Development Framework — 60% dev-time reduction via automated Architectural Compliance + dependency management”
AFTER: “Built an internal LLM-assisted code-review and dependency-graph tool used across 4 squads (~25 engineers); reduced average story cycle time from 6.2 to 2.5 days (measured over Q2–Q3 2025, n=140 stories).”
WHY: Dropped capitalised “Architectural Compliance.” Replaced “60%” with the raw numbers an interviewer will ask for anyway. Adds adoption (4 squads, 25 engineers) — that’s scope signal MAANG needs for L4/E4.

Rewrite 2 — LLM Rate Limiter

BEFORE: “AI Rate Limiter — 40% LLM cost reduction, HA maintained”
AFTER: “Designed and shipped a multi-tenant token-bucket rate limiter in front of OpenAI/Bedrock; cut LLM spend from $X/mo to $Y/mo (~40% drop) while holding p99 added latency under 8 ms across ~Z req/min peak.”
WHY: Absolute spend, latency budget, peak throughput. “HA maintained” is replaced by a concrete latency SLO. Names actual providers — searchable by ATS.

Rewrite 3 — SonarQube / HIPAA

BEFORE: “200+ SonarQube vulnerabilities mitigated → HIPAA compliance, 40% code quality gain”
AFTER: “Remediated 200+ static-analysis findings (47 High / 23 Critical) across the patient-data services, unblocking the HIPAA-aligned audit pass; reduced new-issue inflow by 40% via pre-merge SonarQube quality gates.”
WHY: Honest severity breakdown. “HIPAA compliance” softened to “HIPAA-aligned audit pass” — accurate, defensible. “40% code quality gain” replaced with a concrete inflow metric.

Rewrite 4 — Hibernate optimization

BEFORE: “Hibernate optimization 2.5s → <500ms (80% improvement)”
AFTER: “Cut p95 patient-search latency from 2.5 s to 420 ms by replacing N+1 fetches with batched JPQL + a Redis read-through cache; verified on a 12M-row patient table under 200 req/s synthetic load.”
WHY: Specifies the metric (p95), the technique (batched JPQL + Redis), the dataset (12M rows), and the load (200 req/s). This bullet now stands up to a 20-minute follow-up.

Rewrite 5 — CDC microservice

BEFORE: “CDC microservice — Debezium + Kafka, sub-3-sec sync MongoDB ↔ Postgres”
AFTER: “Built a Debezium → Kafka → Mongo CDC service handling ~N events/min across 12 source tables with end-to-end p95 < 2.8 s; designed idempotent sink with offset checkpointing to survive broker restarts with zero duplicate writes.”
WHY: Adds throughput (events/min), p95 (not “sub-3-sec”), table count, and the hard part of the design (idempotency + offset checkpointing). System-design interviewers love this bullet.

6. Rewritten summary (Sanjay-ready, modular)

Core 3 sentences — use for any audience:

Senior software engineer with 3.5 years building Java/Spring Boot and Python services for clinical-data platforms at Jio Healthcare. Shipped a Debezium-Kafka CDC pipeline (p95 < 3 s), an LLM cost-control rate limiter across multi-tenant traffic, and an FHIR/OpenEHR extraction service consumed by 80+ partner hospitals. Comfortable owning a service end-to-end: design doc → implementation → on-call → metrics.

Swap the middle sentence by audience:

MAANG generalist: “Shipped a Debezium-Kafka CDC pipeline (p95 < 3 s), tuned a JPA hot-path from 2.5 s to 420 ms, and led migration of 20 microservices to LiquiBase with zero-downtime deploys.”
Healthtech-AI: “Shipped a clinical-summarization service over longitudinal EHR data, an LLM rate limiter that cut model spend ~40%, and an FHIR/OpenEHR extraction layer used across 80+ partner hospitals.”
AI-first / infra: “Shipped a multi-tenant LLM rate limiter that cut model spend ~40% at sub-10ms added latency, an internal LLM-assisted dev-tools framework adopted by 4 squads, and a Debezium-Kafka CDC pipeline at p95 < 3 s.”

No “Strategic.” No “Expert.” No “Architecting.” No capitalised abstract nouns. No “millions of users” unless you can name them.

7. Skills section — restructure

Current section reads as a keyword dump. Restructure into Core / Working / Familiar so depth claims match reality.

Core (interview-ready):     Java 17/21, Spring Boot 3, REST, PostgreSQL, Kafka, Debezium,
                            Docker, Kubernetes, FHIR (HL7), OpenEHR
Working (production use):   Python, FastAPI, Redis, MongoDB, Elasticsearch, AWS (EC2, S3),
                            LLM orchestration & prompt engineering, JUnit, Liquibase
Familiar:                   TypeScript, GCP, Argo CD, Jenkins
Tools:                      Git, Postman, SonarQube, Swagger/OpenAPI, IntelliJ

Drop entirely: C++, JavaScript (kept TS only), Django, “Factory Design Pattern,” “Postman Automation,” “System Design (HLD/LLD)” (show it, don’t list it).

Rationale:

An L5 interviewer scanning your skills line decides which deep topic to drill. Every item listed has to survive that drill. Anything you can’t defend in 20 minutes belongs in “Familiar” or off the page.

8. MUST-ADD items he’s missing

Add	Why	Effort
LeetCode profile URL in contact line	Single highest-leverage missing signal for MAANG	5 min, once you have ≥75 problems solved
“Selected Repos” line under GitHub — name 2–3 pinned repos with 1-line each	Recruiters do not browse uninvited	30 min
1 public blog post — recommend “Building a Multi-Tenant LLM Rate Limiter”	Becomes your system-design talking point in 100% of LLM rounds	1 weekend
Portfolio at sanjaydev.online — 3 case studies, each 1 page	Currently a black-box link; either light it up or remove it	1 weekend
Talks / Writing section — internal Jio tech talks count if you can title them	Cheap senior-IC signal	15 min
Concrete scale metrics in bullets — QPS, MAU, GB/day, deploy frequency, MTTR	Backfills the “millions of users” claim with real anchors	1 hour with notes
On-call / incident response — one bullet if you’ve owned a pager	Strong L4/E4 signal at MAANG	20 min

9. ATS scan check — keywords present vs missing

Present (good): Java, Spring Boot, Microservices, Kafka, Debezium, PostgreSQL, MongoDB, Redis, Elasticsearch, Docker, Kubernetes, AWS, CI/CD, FHIR, HL7, LLM, REST, Liquibase, Python, FastAPI, SonarQube, HIPAA.

Missing — add only if genuine:

Keyword	MAANG JD frequency	Add if you can claim
distributed systems	Very high	YES — Debezium pipeline already qualifies
scalability / horizontal scaling	High	YES — soften from “Scalability” buzz to a concrete bullet
gRPC	Medium	Only if you’ve used it; otherwise leave off
GraphQL	Medium	Only if used; healthtech rarely uses it
observability / Prometheus / Grafana / OpenTelemetry	High	Add if your services have dashboards/alerts
SLI / SLO / error budget	High at Google	Add only if you’ve actually defined one
TDD / unit test coverage	Medium	Add coverage % if known
event-driven architecture	High	YES — Kafka work already qualifies, surface the term
idempotency	Medium	Add to the CDC bullet
backpressure / flow control	Medium	Add if true of the rate limiter
p50 / p95 / p99 latency	Very high	Replace generic “sub-3-sec” with “p95 < 2.8 s” everywhere
zero-downtime / blue-green / canary	Medium	LiquiBase bullet already says zero-downtime — good
multi-tenant	High	Rate-limiter bullet should explicitly say “multi-tenant”

Do not keyword-stuff. Each added keyword should map to a real bullet you can defend.

10. 3-month resume action plan

Week 1 — Cut and rewrite

Rewrite the summary using Section 6 template.
Rewrite the 5 bullets in Section 5.
Drop: Octave music app, image colorization, C++, Django, “Strategic.”
Lowercase capitalised abstract nouns site-wide (“Architectural Compliance” → “architecture-compliance checks”).

Week 2 — Create one public proof

Publish 1 blog post: “Building a Multi-Tenant LLM Rate Limiter in Spring Boot” (target 1200–1800 words, Medium or sanjaydev.online).
Add the URL to resume contact line.
Optional: a second post on “Sub-3s CDC with Debezium + Kafka: Idempotency, Offsets, Backpressure.”

Week 3 — Light up GitHub + LeetCode

Make LeetCode profile public; aim for ≥75 problems solved (Medium-heavy).
Pin 3 GitHub repos with proper READMEs (the video-transcoding project is a good candidate).
Add “Selected Repos” line to resume.

Week 4 — A/B test in market

Maintain 2 resume variants: Sanjay_Naik_MAANG.pdf and Sanjay_Naik_HealthAI.pdf (different summary middle sentence + project ordering).
Send 5 referrals/InMails per variant per week. Track callback rate.
After 2 weeks of data, keep the higher-converting variant as the default.

Weeks 5–12 — Compound

DSA: 5–7 LC problems / week. By month 3, ≥150 solved, profile linked.
1 more blog post / month (target: 3 total).
Update resume monthly with any new scale metric you measure at Jio.

11. Final TL;DR

You have an above-average-for-3.5-YoE engineer and a below-average-for-MAANG resume. The work is real — Debezium CDC, an LLM cost-control rate limiter, FHIR/OpenEHR depth, a multithreaded reporting engine across 80+ hospitals, two promotions in three years. None of that is visible to a recruiter doing an 8-second scan, because the resume opens with “Strategic Senior Software Engineer” and never substantiates “millions of users.” Fix the tone, back the numbers, add public proof, and you move from “phone screen sometimes” to “onsite frequently.”

Do these four things this week:

Rewrite the professional summary using the template in Section 6 — drop “Strategic,” “Expert,” “architecting,” “millions of users.”
Rewrite the 5 bullets in Section 5 — replace every % with the absolute number and a measurement method.
Cut Octave, image colorization, C++, Django, and the capitalised abstract nouns.
Add a LeetCode URL to the contact line — even an empty-but-public profile is better than no link, because it signals you’re prepping.

Everything else (blog posts, portfolio, pinned repos, A/B testing) is the next 8 weeks.

Score today: 6.8 / 10 (MAANG SDE-II), 7.5 / 10 (Healthtech-AI). Score achievable in 12 weeks with the plan above: 8.3 / 10 (MAANG SDE-II), 8.8 / 10 (Healthtech-AI).

That’s the difference between “we’ll keep your resume on file” and “can you do an onsite next week.”

Resume Validation — Sanjay Naik vs MAANG SDE-II / Healthtech-AI Bar