JARVIS · ONLINE
--:--:-- PST
N.BINDAL
Console›Module · Subject Dossier
--:--:-- PSTLoaded
[YOU]show me his dossier
› FILE-NB-001·Active·class::alpha
▸ Subject Identification
Nikhil Bindal
Subject · NIKHIL.B
› Designation

Nikhil Bindal

Full-Stack & AI Infrastructure Engineer

Coordinates
37.7749°N · 122.4194°W
San Francisco, CA
› Status
Available for new engagements
› Contact Channels
nikhil.bindal@outlook.com (857) 313-5445
6+
years shipping
AI INFRA
primary domain
12
production projects
OPEN
engagement status
LinkedInGitHubX
▸ Vital Statistics
6+
YEARS SHIPPING
5
COMPANIES
8.4M
DAILY REQUESTS
15+
AI AGENTS
22K
TPS SUSTAINED
73%
LATENCY CUT
▸ Deployed Assets · Reach Stack
PythonFastAPINode.jsPostgreSQLRedisLangGraphCrewAIGPT-4oWebRTCLiveKitKafkaKubernetesGCPAWSQdrant
The journey

Operations Log

Career history rendered as classified mission files. Toggle to view credentials.

Export · Dossier.PDF
▸ Scan Parameters · 5 of 5 ops
› OP-005·Active·class::ALPHA
AI Infrastructure

AI Consultant / AI Infrastructure Engineer

Neurologica · Stealth AI Startup

Sep 2024 → Present
San Francisco, CA
▸ Theatre A · Neurologica — Engineering Contractor
  • Architected a production multi-agent AI coaching platform on GCP Cloud Run combining biometric/HCI signal processing with a persistent personal-memory system, owning system design end to end.
  • Cut end-to-end AI latency 73% (36s → 9.5s) by re-architecting a 6-agent pipeline into a coordinator-orchestrated DAG with prompt consolidation, model tiering, and async parallel execution.
  • Designed "Mnemosyne," a context-scoped memory architecture with semantic retrieval and a temporal-decay lifecycle, bucketed per user and domain so context can never cross boundaries.
  • Engineered Redis-backed concurrency control and session orchestration that prevented state corruption across concurrent real-time voice sessions, plus a parallel voice sidecar (Gemini Live) running sentiment/query/analytics off the hot path.
  • Shipped Python and JavaScript SDKs (32 operations) spanning orchestration, memory, analytics, sessions, and BYOD multi-tenant infrastructure, with CI/CD, audit logging, and webhook delivery.
▸ Theatre B · Stealth AI Startup — AI Solutions Consultant
  • Built "Donna," an AI meeting-intelligence platform, end to end (0→1) automating pre-meeting research, live in-meeting assistance, and post-meeting synthesis, and released it for real users.
  • Designed a 3-phase architecture orchestrating 15 specialized agents behind a uniform agent contract that made parallel orchestration and fan-out trivial to extend.
  • Built hybrid contextual RAG over Qdrant combining vector and keyword retrieval with reciprocal-rank fusion, improving retrieval relevance ~40% over a naive dense baseline (measured, not estimated).
  • Implemented a real-time voice pipeline (LiveKit WebRTC + Deepgram STT + Cartesia TTS) targeting sub-200ms perceived latency, with a replica-independent WebSocket fan-out over Redis pub/sub.
  • Built "RecoMe," a personal interest-graph & agentic recommendation engine — a capture → signal → graph → agent → surface pipeline turning cross-platform activity (15+ sources) into a typed Neo4j interest graph plus Qdrant per-user vectors, with recommendations streamed to the client over SSE.
  • Orchestrated 5 background agents behind a single Guardian gate enforcing a hard $0.10/user/day LLM cost cap, throttling, and quiet hours, with a prompt-injection-resistant scorer (the LLM writes prose only, never the verdict); ran on BullMQ workers with idempotent Stripe metering and a dual consumer + multi-tenant surface on one backend (TypeScript, Hono, Prisma).
▸ Deployed Assets
PythonFastAPINode.jsTypeScriptHonoPostgreSQLRedisQdrantNeo4jWebRTCLiveKitGemini LiveCrewAILangGraphGPT-4oDeepgramCartesiaBullMQStripeKubernetesGCP Cloud RunPrometheus
› OP-004·Concluded·class::BETA
Full-Stack Engineering

Full-Stack & AI Engineer

Northeastern University — Minkara Computational Lab

Feb 2023 → Dec 2023
Boston, MA
▸ Mission Outcomes
  • Built a semantic-search & visualization platform for biomedical research (FastAPI, PostgreSQL + pgvector, React, D3.js) letting researchers search 10K+ papers and molecular datasets by meaning via embeddings and vector search.
  • Built AWS Batch distributed simulation pipelines orchestrating thousands of parallel molecular computations and cutting compute cost ~40% via spot instances, right-sizing, and scale-to-zero with checkpoint/restart.
  • Owned backend for scientific data ingestion, metadata APIs, and ML-inference workflows serving models as isolated services so heavy inference never blocked the API.
  • Designed accessibility tooling for visually impaired researchers — tactile-graphics generation from molecular coordinates and screen-reader/sonification integrations, validated with blind users including the lab PI.
▸ Deployed Assets
FastAPIPostgreSQLpgvectorReactD3.jsAWS BatchPythonTypeScriptDocker
› OP-003·Concluded·class::BETA
Distributed Backend

Software Engineer

Times Internet — TOI+ subscription platform

Apr 2021 → Jul 2022
Noida, India
▸ Mission Outcomes
  • Built backend for TOI+ serving ~8.4M daily requests — Node/Express subscription & paywall services with JWT auth and Redis-cached entitlements (sub-ms checks), keeping origin load low by offloading cacheable content to the Akamai edge.
  • Designed a Verdaccio-based micro-frontend system packaging shared UI as versioned internal npm widgets, so 70+ city portals consumed updates by version bump instead of duplicated code — publish cadence decoupled from consume cadence.
  • Migrated 70+ legacy XML/XSLT portals to React micro-frontends via a strangler-fig rollout (old + new in parallel, per-portal feature flags, stable API contracts), reaching ~92/100 Lighthouse with instant rollback.
  • Built the real-time Kafka event pipeline feeding the Signals personalization engine (behavior events → user feature profiles, ~90s refresh), contributing to a measured ~9% CTR lift on recommendations.
▸ Deployed Assets
Node.jsExpressRedisKafkaReactVerdaccioJWTAkamaiDocker
› OP-002·Concluded·class::BETA
Distributed Backend

Founding Software Engineer

Progcap — collateral-free SMB lending fintech

Jan 2019 → Mar 2021
New Delhi, India
▸ Mission Outcomes
  • Owned the underwriting & transaction backbone of a collateral-free lending platform — Node/Express microservices on an event-driven Kafka backbone, with PostgreSQL as the transactional system of record (+ append-only ledger) and MongoDB for high-write capture.
  • Cut decision latency 90% (8.7s → 890ms) by parallelizing serial KYC/fraud/credit-score checks, trimming the hot path, and adding compound indexes plus Redis caching with connection pooling.
  • Integrated XGBoost credit scoring on alternative data as an isolated service with timeouts and rule-based fallback, reducing false negatives ~19% (model + rules, measured on repayment cohorts).
  • Engineered effectively-once disbursement via idempotency — guarded atomic state transitions, idempotency-keyed bank/NPCI calls, and partitioned Kafka consumers; load-tested to ~22K events/sec at the event tier.
  • Built a Python/Celery service for feature assembly, batch jobs, and reconciliation against the append-only ledger as the source of truth.
▸ Deployed Assets
Node.jsExpressKafkaPostgreSQLMongoDBRedisXGBoostPythonCelery
› OP-001·Concluded·class::BETA
Full-Stack Engineering

Software Engineer

LiveMedia / LiveChek — insurance telematics

Aug 2017 → May 2018
New Delhi, India
▸ Mission Outcomes
  • Built backend services and APIs for real-time telematics and behavioral analytics processing driving-behavior and user-interaction streams for insurance workflows.
  • Contributed to early-stage product and architecture decisions in a fast-moving startup environment.
▸ Deployed Assets
Node.jsPythonPostgreSQLREST APIsTelematics
›Open a channel→
How I work

Doctrine

Six operating principles drawn from staff-level engineering and founding-team experience — the rules I default to when there's no playbook.

▸ Doctrine · 01 · craft

Production-First

I build with the assumption a system will be paged at 3am. Observability, idempotency, and graceful degradation are designed in — not bolted on after the first incident.

▸ Doctrine · 02 · vision

AI as Leverage

AI is not a replacement for engineering judgment — it's leverage. I build agentic systems that compound human decisions and keep a human in the loop on every escape hatch.

▸ Doctrine · 03 · velocity

Ship, Then Polish

I optimize for measurable outcomes over premature abstractions. Get a working version into production, instrument it, then iterate on what the data actually shows — not what the design doc predicted.

▸ Doctrine · 04 · ownership

Ownership Over Tasks

I own outcomes, not assignments. If something blocks the goal — a broken test in another service, an unclear product spec, a vendor outage — it becomes my problem until it's resolved or properly escalated.

▸ Doctrine · 05 · judgment

Strong Opinions, Loose Grip

I form sharp views early and argue for them with evidence. The moment better evidence shows up, I drop the position cleanly. Disagree hard, commit fast — politics costs more than humility ever does.

▸ Doctrine · 06 · leverage

Force Multiplier

My code is only half the job. Clear APIs, sharp docs, sturdy tests, and unblocking teammates compound my output beyond what any individual contributor can ship — that's how staff-level impact actually shows up.

◇ end of dossier · file sealed